v0.15.2: changelog and README

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

BACKLOG.md

@@ -30,17 +30,31 @@ What is left, most needed first. The detail is in the item each names.
 **Soon: the customer's experience, and HA.**
 4. ~~**C24**~~: done 2026-10-21. The store's, catalog's and mirror's errors, percent signs, footer times and stale
    forms are in the page's language. Left: the /docs documents, and a language kept per person (cookie only today).
-5. **J4**: the model mirror on more than one machine, the last single point of an HA install. (M)
+5. **J4**: the model mirror on more than one machine. Code done 2026-10-29: mirrors kept in step (`mirror serve
+   --peer`, removals carried), `mirror.peers` and failover in the control plane, nodes given the file on a mirror that
+   answers, `mirror add-peer` and the setup's second mirror. Tried 2026-10-29 on two VMs:
+   - `add-peer` set the second up over ssh;
+   - a file came across in ~30 s, checksum matching;
+   - a version added on the second while the first was off reached the first when it came back;
+   - one removed on the first while the second was off was removed there when it came back, not copied back.
+
+   Left: the same with a control plane and a node, a model served while the first mirror is off. That part (failover
+   in the control plane, the node given the other mirror's file) is so far tried only in tests.
 6. **F5**: large imports limited by the embedding engine's slots. (M)
 
-**From Kong** (section L): L1 done; L9, then L2 and L3, then L4/L5. L6 and L7 are llmgw-80's.
+**From Kong** (section L): L1–L5 and L9 done (v0.15.0); left of L4 is routing by cost between outside
+providers, of L5 cache-write and audio prices. L6 and L7 are llmgw-80's; L8 (A2A) when asked.
+
+**Models we can trust** (section M, 2026-10-25): M1 done (pickles refused; the explorer lists only what this
+cluster's hardware holds, with CPU and GPU tok/s). Next M2 (every download pinned and checked against Hugging
+Face), then M5 (engines without the internet), M3, M4, M6.
 
 **Governance.**
 7. **C1**: a tenant-admin role, and forms hidden from viewers.
 8. **Audit before and after**: what a change changed, and changes made with the CLI recorded too.
 
 **When there is room.**
-9. **I1**: answers to similar questions from the cache. (M)
+9. ~~**I1**~~: done 2026-10-25 (in memory, not pgvector); left: the 40-rewordings run on the PoC.
 10. **H1**: tried on the PoC.
 11. **C4**: Parquet, and a scheduled push for usage export.
 
@@ -59,7 +73,7 @@ of several control planes. 15. **J6**: rate limits per process (a note).
 **Not code, the operator's to decide or do:**
 - Where Alertmanager sends (its receiver is "none"): with a judge down, nobody hears.
 - The fabric test for K.
-- The PoC's control planes run a dev build of main (v0.14.9-11-g4eb49d0); its nodes run v0.14.8.
+- The PoC runs a dev build of main (v0.15.0-7-ga583359) on its control planes and nodes, the mirror v0.15.0.
 
 ---
 
@@ -1784,6 +1798,63 @@ A customer that already runs Kong puts llmgw behind it: `ai-proxy` with
 route. A page in docs/ with the configuration, tried against Kong 3.9 in a
 container, and what Kong's token accounting sees.
 
+## M. Models from Hugging Face we can trust (written 2026-10-25)
+
+The user asked how we know a model fetched from Hugging Face is safe. Two
+questions: whether its files can run code on our machines, which we can
+close, and whether the model itself was trained to misbehave, which no scan
+proves — only who published it, an eval before it serves, and the
+guardrails narrow that. What existed: `--trust-remote-code` refused, no
+repository Python fetched, ComfyUI nodes pinned to a commit, engines
+hash-pinned and run as `llmgw-engine`, mirror versions pinned to a commit
+and never overwritten. What did not: everything below.
+
+### M1. Weights in a pickle format refused (S) — done 2026-10-25
+`.bin`, `.pt`, `.pth`, `.ckpt`, `.pkl` run code as they are loaded, which is
+how the malicious models found on Hugging Face work. The node refuses such a
+source, or a mirror set holding one, before fetching anything; a model being
+added or given such a source is refused; vLLM and SGLang downloading a model
+by its id are given `--load-format safetensors`. The explorer hides a repository with neither GGUF nor safetensors
+(under "runs here", ticked by default) and does not offer a repository's
+pickle files. `allow_pickle: true` on a model lets it through, on purpose. On the
+PoC, `triposr` (`hf:stabilityai/TripoSR/model.ckpt`) is the one model that
+needs it.
+
+### M2. Every download pinned and checked against Hugging Face (M)
+- `hf:org/repo/file` follows `main` and is checked by nothing; a bare
+  `org/repo` given to vLLM is downloaded by vLLM, of whatever `main` is.
+  Both: resolve to the commit when the model is added, keep it in the
+  catalog's origin, and fetch that commit only (`--revision` for vLLM).
+- The sha256 Hugging Face publishes for each LFS file compared on every
+  path, and a file that differs removed: `mirror add`, the manager, the
+  node's `hf:` and http sources, mirror set files with no sha recorded.
+- Done when: a file changed on Hugging Face after a model was added is
+  refused by the node with both checksums named.
+
+### M3. Hugging Face's own scan (S)
+Hugging Face scans every file (pickle imports, ClamAV, partners' scanners)
+and marks a file unsafe. Ask for it with the repository listing; refuse an
+unsafe file, show the result on the mirror page and in the catalog's origin.
+The API field has to be confirmed first.
+
+### M4. Publishers we trust, and the rest asked about (S)
+An allow-list of organisations in the cluster's settings (Qwen, meta-llama,
+mistralai, google, microsoft, …). A model from anyone else is added only by
+an admin who says so, and who said it is kept in the origin. The PoC already
+serves four community models whose training nobody here checked.
+
+### M5. Engines without the internet (M)
+The firewall is inbound only, so an engine that was made to run code can
+send what it reads anywhere. With weights from the mirror the engines need
+no egress: an nft output rule for the engine user (allow the mirror, the
+cluster, loopback), systemd hardening of the agent's engines
+(`NoNewPrivileges`, `ProtectSystem`, private /tmp). Agent itself still root.
+
+### M6. An eval before a new model serves (S)
+`eval` and `canary` exist; make a new model from an untrusted publisher
+(M4) serve only after its eval suite passes, so a model trained to answer
+wrongly on a trigger has at least been asked the suite.
+
 ## D. Facts worth knowing before starting
 
 - Admin API: `127.0.0.1:9090`, behind the token in

CHANGELOG.md

@@ -19,12 +19,234 @@ is only released once its **exit criterion** is met, not when its code lands.
 | v0.12.0 | — | The cluster's settings in the store, in force when changed; the firewall switch; wipe | released |
 | v0.13.0 | — | The whole system stopped and started from one command; Qwen3Guard judges, on a GPU and CPUs together | released |
 | v0.14.0 | — | Knowledge bases: documents from people's own machines, kept apart per tenant, answered from | released (v0.14.1: adapters, the tree over the API, the firewall on every machine; v0.14.2: pictures from ComfyUI; v0.14.3: 3D models as jobs; v0.14.4: any ComfyUI workflow as a job, presets, licences; v0.14.5: picture and 3D models that add themselves; v0.14.6: setup that explains every part and lets it be chosen, the gateway on all three HA machines; v0.14.7: the judge sized for its machine; v0.14.8: sign-in with Active Directory, several keys for one outside provider; v0.14.9: sign-in set up from the console, a node's console that shows the machine) |
-| v0.15.0 | — | The company's MCP tools and VS Code through the gateway; guardrails that ask outside services and put redacted values back; answers to similar questions | released |
+| v0.15.0 | — | The company's MCP tools and VS Code through the gateway; guardrails that ask outside services and put redacted values back; answers to similar questions | released (v0.15.1: Strata, a 125B model of experts on one card; a second mirror kept in step; pickled weights refused; a served model's context changed; v0.15.2: a served model moved to more cards) |
 | v1.0.0 | 5 | What phase 5 has left: precise KV routing proven on vLLM, the block cost picker, defragmentation across NVLink domains, prefill/decode disaggregation | planned |
 
 Until v1.0.0 the minor version is the breaking one: configuration formats and Go
 APIs change between minors without a deprecation period.
 
+## v0.15.2 — 2026-10-29
+
+A served model moved to more cards, or fewer, while it is served.
+
+**Upgrading.**
+- No migrations, and nothing changes until the command is used. The control
+  planes carry it; the nodes need nothing new.
+
+### Added
+
+- **A served model moved to more cards, or fewer.** `llmgw ctl model gpus
+  MODEL N [--profile P] [--in-place]`, `POST /api/v1/models/{name}/gpus`, and
+  GPUs… in the console's model menu. How many cards of one node each replica
+  uses: tensor parallel for vLLM, SGLang and TensorRT-LLM, checked against
+  the model's KV heads, and the layers split for llama.cpp. Refused for a
+  count no node has, or whose cards would not hold the model and its
+  context, with the figures; its replicas are then started again on it, one
+  by one, beside the old — or, with `--in-place` when there is no room
+  beside them, each stopped first.
+
+## v0.15.1 — 2026-10-29
+
+A 125B model on one card, a second mirror, and weights that run code
+refused. Strata runs as an engine: a mixture-of-experts model with its
+experts in RAM, served from one 24 GB card. A second model mirror keeps in
+step with the first, so an HA install has no single machine left. Pickled
+weights are refused unless a model says otherwise. A served model's context
+can be changed in place. The explorer says how fast a model would answer
+here before it is fetched, and a card in the wrong PCIe slot is said so.
+
+**Upgrading.**
+- No migrations. Upgrade the control planes before the nodes, for the
+  pickle refusal below.
+- A model whose weights are in a pickle format (`.bin`, `.pt`, `.pth`,
+  `.ckpt`, `.pkl`) stops being placed: its node refuses the source. Give it
+  its safetensors or GGUF, or add it again with `allow_pickle: true`
+  (`llmgw ctl model add -f`) if its files were checked. Look at the
+  models' sources before upgrading the nodes: one ending in
+  `.ckpt` or `.bin`, or a mirror version holding `pytorch_model.bin`. Upgrade the control planes first: a node on
+  this version refuses those weights until its control plane can say
+  `allow_pickle`.
+- Strata is opt-in: nothing installs it until `engines` or an
+  `engine_overrides` entry names a version (`0.2.7`).
+
+### Added
+
+- **Strata as an engine: a 125B model of experts on one card.** Strata
+  (github.com/Niko1221/Strata) keeps a model's experts in RAM and the ones
+  used most in VRAM. On an RTX 6000 24 GB with 251 GB of RAM it serves
+  Qwen3.8-Flash-Next 125B (IQ2_XS) at 70–87 tokens a second to one client;
+  more clients share about that, its slots buying fairness rather than
+  throughput on one card. It runs a few families only — Qwen3.8-Flash-Next
+  and its variants, Qwen3.6-35B-A3B, DeepSeek V4.1-Flash — each named in a
+  profile's args with `--family` and `--size`
+  (`--family qwen --size IQ2_XS --context 32768`). The source is the
+  directory of the model's GGUF files, or the repository, which Strata then
+  downloads itself, pinned. The agent installs it on an NVIDIA node of
+  x86-64: its source pinned by digest, in a Python environment, and its
+  release's engine checked against the digest GitHub publishes — the CUDA 13
+  build for Turing and newer on driver 580, the CUDA 12.9 build for Pascal
+  to Ada, a V100 included, on driver 575. A replica prepares its model the
+  first time (minutes; its draft layer is a 5 GB download) and loads in
+  about a minute and a half after. One card of one node per replica, for
+  now; not in an offline bundle, its packages coming from PyPI.
+
+- **A served model's context, changed.** `llmgw ctl model context MODEL
+  TOKENS [--slots N]`, `POST /api/v1/models/{name}/context`, and Context
+  length… in the console's model menu. The catalog, the engine's arguments and
+  the memory change together, checked against what the model was trained for
+  and what its cards hold; its replicas are then started again one by one. The
+  models API shows `context_length` and `max_output`.
+
+- **A second model mirror, kept in step.** In a three-machine install the setup screens put one on a database machine
+  (`--second-mirror 1|2`). An install with one mirror gets one with `sudo llmgw mirror add-peer user@host`.
+  - **In step:** each mirror reads the other every thirty seconds. It copies what it lacks, checksum checked, and
+    removes what the other removed, even if it was down at the time (`llmgw mirror serve --peer`).
+  - **Control planes:** they name both (`mirror.peers`) and use whichever answers.
+  - **Nodes:** while one mirror is down, a node is given a model's file on the other.
+
+  This takes away the last single point of an HA install: with the first machine off, models are still added and
+  fetched.
+
+### Fixed
+
+- **An engine that computes in a process of its own group was not its
+  replica's.** A GPU replica read as running on the CPU, and the memory it
+  held on its card as somebody else's, which placement then kept free of
+  other work. A process the replica started, at any remove, is now the
+  replica's. And what a replica computes on, once sent, was never sent
+  again: a wrong reading kept from before an agent's restart stayed wrong.
+  It is sent again when it changes, and read again when a GPU replica was
+  kept as the CPU.
+
+- **VS Code's sign-in lost its place through the sign-in page**, so the
+  prompt to confirm came every time; a link to the same gateway by plain
+  http could take an https setting down to http; and a new build of the
+  extension was not installed over an old one (the command now says
+  `--force`).
+- **The install guide's recordings were not checked for secrets**, only
+  the page; every frame is now.
+- **MCP server fields could not be emptied**, an address could carry a
+  password, the cluster's own server could not be changed from the
+  console, and a grant to a key that does not exist was said done.
+- **A guardrail service two policies named was asked as the first said**,
+  not the stricter; a `null` answer read as safe; and with `restore` the
+  request log kept the answer with the redacted values back in it.
+- **What an engine downloads by a Hugging Face id was not held to the
+  pickle refusal** — an adapter by its id, a TensorRT-LLM model's; a model
+  with either now needs `allow_pickle`.
+- **`server init` run again on a machine already set up was refused its
+  memory**, as on an 8 GB install fixing its certificate with `--host`, and
+  judges that would not be installed (no mirror) were counted.
+- **Removing an MCP server left its sealed credential in the store**, and
+  replacing a credential under another header kept the old one sent. A
+  server's secrets now go when it no longer reads them.
+- **A cloned repository could point the VS Code extension at another
+  gateway.** `llmgw.gatewayUrl` is a machine setting now, never taken from a
+  workspace.
+- **A node's memory bandwidth could be measured in its cache** on a busy
+  node, and was 0 after a reconnect until the next heartbeat. The buffer is
+  at least eight times the processor's last cache.
+- **A project kept in the cluster could have its questions embedded outside
+  it.** The similar-question cache now asks the embedding model as the
+  question's project would, external fallbacks refused, and skips a model
+  that is itself outside. Its comparing no longer holds the cache's lock,
+  which every project's lookups waited on.
+- **Restored values were left as placeholders in streamed tool calls**, in
+  completions and in reasoning, and a value with a quote in a whole answer's
+  tool-call arguments broke their JSON.
+- **An outside guardrail service read what `redact` had taken out**, and a
+  streamed answer was sent to it twice. The docs now say a service stops a
+  streamed answer only under `stream: buffer_all`.
+- **A replica found slow once could stay unused.** Its time to first token
+  is forgotten after five minutes without an answer, so it is tried again.
+- **Weights that run code got past the refusal in four ways.** A mirror
+  version already on a node was used without being looked at again; a
+  `?query` or `#fragment` on an `hf:` source hid the file's name; `.pt2` was
+  not on the list; and a profile could write `--load_format`,
+  `--trust_remote_code` or an abbreviation such as `--trust-remote`, which
+  vLLM and argparse read as the flag, or give SGLang `--load-format`. All
+  refused now. TripoSR's preset says `allow_pickle`, without which it could
+  not be added, and a shared ComfyUI environment takes its first member's.
+- **A temperature written as a string was refused.** Since the GenAI traces
+  read it, `"temperature": "0.7"` got a 400 before reaching the engine; it
+  is read loosely again, for the trace only.
+- **A reused answer could carry another caller's personal data.** With
+  `restore: true` and the answer cache on, two callers' requests that
+  redacted to the same text shared one answer — with the first caller's
+  address put back in it. Such requests are now never answered from the
+  cache nor kept in it.
+- **Granting a server's tools could take others away; revoking could close
+  a level to every tool.** A level with no tool list sees what the levels
+  above grant; giving it its first list made it see only that list. And a
+  level left with an empty list was closed to every tool, which the screens
+  did not show. Now a first list starts from the nearest list above, taking
+  a server away leaves the rest or goes back to no list, the screens say
+  when a level above keeps granted tools out of reach and which levels are
+  closed, and the console reads and writes the level at once rather than
+  from what it loaded.
+- **Similar questions about different pictures got each other's answer.**
+  A question with a picture, a sound or a file in it is reused only when it
+  is the same request.
+
+### Added
+
+- **A card in the wrong PCIe slot is said so.** Each agent reads how every
+  card is linked — the fastest generation card and slot allow together, and
+  the lanes in use — against the card's own most, from the kernel (the
+  driver's figure is what card and slot allow together); a card on fewer lanes or
+  an older generation is a node warning, with what it carries and what it
+  could, on the dashboard, the console and in `llmgw doctor`.
+- **And on a web page, `/admin/mcp`.** For administrators: the same list,
+  adding a server with its credential sealed, testing it and listing its
+  tools, granting them in the tree beside that level's other grants and
+  limits, revoking, changing, replacing the credential, turning it off,
+  removing; and the line that points an agent at this gateway's `/mcp`.
+  In English and Turkish; every change in the audit log, the secret not.
+- **The company's MCP servers in the console.** A screen of its own: each
+  server with its state, who has its tools, its rate, the names of the
+  credentials it is sent and its address; add one with a bearer token or
+  a header kept sealed, test it and list its tools, grant them to a team,
+  project, organization, person or key beside that level's other grants,
+  revoke, change, replace the credential, turn off, remove.
+- **And on the cards, as this cluster would lay it out.** The explorer's
+  **GPU tok/s** and a repository's files give one reply's speed for each
+  layout the planner offers here — one card, tensor parallel across a
+  node's cards, llama.cpp's layer split, a pipeline across nodes on a
+  fabric — the fastest of them, with the layout in the tooltip. From each
+  card's memory bandwidth by its model name, NVLink or PCIe between a
+  node's cards, and InfiniBand or RoCE between nodes; a cluster of mixed
+  classes is judged class by class. A model of experts reads 1.6 times its
+  active share, as Qwen3-Coder-30B-A3B did on cuda01's RTX 6000 (144
+  tok/s; qwen3-vl-8b Q8: 62).
+- **How fast a model would answer on a CPU, before it is fetched.** Each
+  agent measures how fast its node's memory reads when it starts (`Host`
+  gets `mem_bandwidth_gbs`, the node's view `mem_bandwidth_gbs`); the
+  explorer's **CPU tok/s** column and a repository's GGUF files say about
+  how many tokens a second one reply gets on the fastest node that holds
+  the model — the bandwidth over the bytes a token reads, its active part
+  for a model of experts. Calibrated on the PoC against llama.cpp's own
+  timings: within 15% from 1.5B to 14B.
+- **The explorer lists what this cluster can run.** Judged by the nodes
+  as they are now — every card's memory and compute capability, every
+  node's RAM — through the same planner as serving a model, rather than a
+  size picked from a list. A model too big for any node, in FP8 without a
+  card of 8.9, in MLX, or in neither GGUF nor safetensors, is left out and
+  counted; the page says what it judged by. With cards, safetensors repositories are
+  listed beside GGUF. On a repository's page a file too big for the
+  cluster is shown with what it needs and no Fetch button.
+- **Weights that run code are refused.** Python pickles — `.bin`, `.pt`,
+  `.pth`, `.ckpt` — run whatever they hold when loaded, which is how the
+  malicious models on Hugging Face work. The node refuses such a source,
+  or a mirror version with one in it, before fetching it; adding such a
+  model is refused; vLLM and SGLang downloading a model by its id read its
+  safetensors only; the mirror page marks the file. `allow_pickle: true`
+  on a model lets it through on purpose. The mirror's explorer no longer
+  lists a repository with nothing a node loads (no GGUF, no safetensors)
+  while "runs here" is ticked, and a repository's pickle files are not
+  offered, only counted. BACKLOG M lists what is next:
+  checksums against Hugging Face everywhere, its own scan, trusted
+  publishers, engines without the internet.
+
 ## v0.15.0 — 2026-10-25
 
 The company's tools and editors through the gateway, and guardrails that

README.md

@@ -41,7 +41,7 @@ Releases track the roadmap phases; see [CHANGELOG.md](CHANGELOG.md).
 | v0.12.0 | — | The cluster's settings in the store, in force when changed; the firewall switch; wipe | released |
 | v0.13.0 | — | The whole system stopped and started from one command; Qwen3Guard judges, on a GPU and CPUs together | released |
 | v0.14.0 | — | Knowledge bases: documents from people's own machines, kept apart per tenant, answered from | released (v0.14.1: adapters, the tree over the API, the firewall on every machine; v0.14.2: pictures from ComfyUI; v0.14.3: 3D models as jobs; v0.14.4: any ComfyUI workflow as a job, presets, licences; v0.14.5: picture and 3D models that add themselves; v0.14.6: setup that explains every part and lets it be chosen, the gateway on all three HA machines; v0.14.7: the judge sized for its machine; v0.14.8: sign-in with Active Directory, several keys for one outside provider; v0.14.9: sign-in set up from the console, a node's console that shows the machine) |
-| v0.15.0 | — | The company's MCP tools and VS Code through the gateway; guardrails that ask outside services and put redacted values back; answers to similar questions | released |
+| v0.15.0 | — | The company's MCP tools and VS Code through the gateway; guardrails that ask outside services and put redacted values back; answers to similar questions | released (v0.15.1: Strata, a 125B model of experts on one card; a second mirror kept in step; pickled weights refused; a served model's context changed; v0.15.2: a served model moved to more cards) |
 | v1.0.0 | 5 | What phase 5 has left: precise KV routing proven on vLLM, the block cost picker, defragmentation across NVLink domains, prefill/decode disaggregation | planned |
 
 As of v0.6.0 a cluster runs unattended: agents report their hardware, the

api/gen/llmgw/v1/agent.pb.go

@@ -1206,9 +1206,12 @@ type Host struct {
 	CpuFlags []string `protobuf:"bytes,5,rep,name=cpu_flags,json=cpuFlags,proto3" json:"cpu_flags,omitempty"`
 	// nvidia_driver is the driver version, empty when there is none. A card
 	// without a driver is not a GPU anything can use.
-	NvidiaDriver  string `protobuf:"bytes,6,opt,name=nvidia_driver,json=nvidiaDriver,proto3" json:"nvidia_driver,omitempty"`
-	unknownFields protoimpl.UnknownFields
-	sizeCache     protoimpl.SizeCache
+	NvidiaDriver string `protobuf:"bytes,6,opt,name=nvidia_driver,json=nvidiaDriver,proto3" json:"nvidia_driver,omitempty"`
+	// mem_bandwidth_gbs is how fast the memory reads, in GB/s, measured when
+	// the agent started: what decides a model's speed on the CPU. 0 unknown.
+	MemBandwidthGbs float64 `protobuf:"fixed64,7,opt,name=mem_bandwidth_gbs,json=memBandwidthGbs,proto3" json:"mem_bandwidth_gbs,omitempty"`
+	unknownFields   protoimpl.UnknownFields
+	sizeCache       protoimpl.SizeCache
 }
 
 func (x *Host) Reset() {
@@ -1283,6 +1286,13 @@ func (x *Host) GetNvidiaDriver() string {
 	return ""
 }
 
+func (x *Host) GetMemBandwidthGbs() float64 {
+	if x != nil {
+		return x.MemBandwidthGbs
+	}
+	return 0
+}
+
 // Storage is one filesystem a node keeps weights on.
 //
 // Per filesystem rather than per configured directory: free space belongs to
@@ -1579,7 +1589,14 @@ type GPU struct {
 	// is then 1000 + 100 × card + device, uuid the instance's (MIG-…), and
 	// memory the instance's share; utilization_percent stays zero, because no
 	// driver measures it per instance.
-	Mig           *MIGDevice `protobuf:"bytes,22,opt,name=mig,proto3" json:"mig,omitempty"`
+	Mig *MIGDevice `protobuf:"bytes,22,opt,name=mig,proto3" json:"mig,omitempty"`
+	// How the card is linked to its host: the fastest generation card and
+	// slot allow together, the card's own fastest, and the lanes in use and
+	// the card's most. Zero when the driver did not say.
+	PcieGen       int32 `protobuf:"varint,23,opt,name=pcie_gen,json=pcieGen,proto3" json:"pcie_gen,omitempty"`
+	PcieGpuGen    int32 `protobuf:"varint,24,opt,name=pcie_gpu_gen,json=pcieGpuGen,proto3" json:"pcie_gpu_gen,omitempty"`
+	PcieWidth     int32 `protobuf:"varint,25,opt,name=pcie_width,json=pcieWidth,proto3" json:"pcie_width,omitempty"`
+	PcieWidthMax  int32 `protobuf:"varint,26,opt,name=pcie_width_max,json=pcieWidthMax,proto3" json:"pcie_width_max,omitempty"`
 	unknownFields protoimpl.UnknownFields
 	sizeCache     protoimpl.SizeCache
 }
@@ -1768,6 +1785,34 @@ func (x *GPU) GetMig() *MIGDevice {
 	return nil
 }
 
+func (x *GPU) GetPcieGen() int32 {
+	if x != nil {
+		return x.PcieGen
+	}
+	return 0
+}
+
+func (x *GPU) GetPcieGpuGen() int32 {
+	if x != nil {
+		return x.PcieGpuGen
+	}
+	return 0
+}
+
+func (x *GPU) GetPcieWidth() int32 {
+	if x != nil {
+		return x.PcieWidth
+	}
+	return 0
+}
+
+func (x *GPU) GetPcieWidthMax() int32 {
+	if x != nil {
+		return x.PcieWidthMax
+	}
+	return 0
+}
+
 // MIGDevice is which card a MIG instance is carved from, and how.
 type MIGDevice struct {
 	state       protoimpl.MessageState `protogen:"open.v1"`
@@ -2943,8 +2988,12 @@ type Heartbeat struct {
 	// is not handed out. Empty from an agent that does not measure it, or from
 	// one that could not this time.
 	GpuForeignMemoryMb []int32 `protobuf:"varint,19,rep,packed,name=gpu_foreign_memory_mb,json=gpuForeignMemoryMb,proto3" json:"gpu_foreign_memory_mb,omitempty"`
-	unknownFields      protoimpl.UnknownFields
-	sizeCache          protoimpl.SizeCache
+	// mem_bandwidth_gbs is how fast the node's memory reads, in GB/s, once
+	// the agent has measured it — in the background after it starts, so a
+	// registration usually has none yet. 0 means not measured.
+	MemBandwidthGbs float64 `protobuf:"fixed64,20,opt,name=mem_bandwidth_gbs,json=memBandwidthGbs,proto3" json:"mem_bandwidth_gbs,omitempty"`
+	unknownFields   protoimpl.UnknownFields
+	sizeCache       protoimpl.SizeCache
 }
 
 func (x *Heartbeat) Reset() {
@@ -3110,6 +3159,13 @@ func (x *Heartbeat) GetGpuForeignMemoryMb() []int32 {
 	return nil
 }
 
+func (x *Heartbeat) GetMemBandwidthGbs() float64 {
+	if x != nil {
+		return x.MemBandwidthGbs
+	}
+	return 0
+}
+
 // Provision is the software a node should have: which engines, at which
 // versions. Nothing more.
 //
@@ -3790,7 +3846,10 @@ type EngineProfile struct {
 	Adapters []*LoraAdapter `protobuf:"bytes,12,rep,name=adapters,proto3" json:"adapters,omitempty"`
 	// kv_events has a vLLM engine publish its KV cache events to the agent, for
 	// the router's precise index.
-	KvEvents      bool `protobuf:"varint,13,opt,name=kv_events,json=kvEvents,proto3" json:"kv_events,omitempty"`
+	KvEvents bool `protobuf:"varint,13,opt,name=kv_events,json=kvEvents,proto3" json:"kv_events,omitempty"`
+	// allow_pickle lets the node fetch weights in a pickle format, which run
+	// code when loaded: the catalog's allow_pickle, set on purpose.
+	AllowPickle   bool `protobuf:"varint,14,opt,name=allow_pickle,json=allowPickle,proto3" json:"allow_pickle,omitempty"`
 	unknownFields protoimpl.UnknownFields
 	sizeCache     protoimpl.SizeCache
 }
@@ -3916,6 +3975,13 @@ func (x *EngineProfile) GetKvEvents() bool {
 	return false
 }
 
+func (x *EngineProfile) GetAllowPickle() bool {
+	if x != nil {
+		return x.AllowPickle
+	}
+	return false
+}
+
 // RemoveWeights deletes cached model files a node no longer needs.
 //
 // The paths are computed by the control plane from the catalog, but the agent
@@ -5180,7 +5246,7 @@ const file_llmgw_v1_agent_proto_rawDesc = "" +
 	"\x05value\x18\x02 \x01(\tR\x05value:\x028\x01\x1aA\n" +
 	"\x13EngineVersionsEntry\x12\x10\n" +
 	"\x03key\x18\x01 \x01(\tR\x03key\x12\x14\n" +
-	"\x05value\x18\x02 \x01(\tR\x05value:\x028\x01\"\xb1\x01\n" +
+	"\x05value\x18\x02 \x01(\tR\x05value:\x028\x01\"\xdd\x01\n" +
 	"\x04Host\x12\x13\n" +
 	"\x05os_id\x18\x01 \x01(\tR\x04osId\x12\x1d\n" +
 	"\n" +
@@ -5189,7 +5255,8 @@ const file_llmgw_v1_agent_proto_rawDesc = "" +
 	"osCodename\x12\x12\n" +
 	"\x04arch\x18\x04 \x01(\tR\x04arch\x12\x1b\n" +
 	"\tcpu_flags\x18\x05 \x03(\tR\bcpuFlags\x12#\n" +
-	"\rnvidia_driver\x18\x06 \x01(\tR\fnvidiaDriver\"u\n" +
+	"\rnvidia_driver\x18\x06 \x01(\tR\fnvidiaDriver\x12*\n" +
+	"\x11mem_bandwidth_gbs\x18\a \x01(\x01R\x0fmemBandwidthGbs\"u\n" +
 	"\aStorage\x12\x12\n" +
 	"\x04path\x18\x01 \x01(\tR\x04path\x12\x1d\n" +
 	"\n" +
@@ -5219,7 +5286,7 @@ const file_llmgw_v1_agent_proto_rawDesc = "" +
 	"leaderPort\x12\x1d\n" +
 	"\n" +
 	"profile_fp\x18\x0f \x01(\tR\tprofileFp\x12+\n" +
-	"\acompute\x18\x10 \x01(\v2\x11.llmgw.v1.ComputeR\acompute\"\xca\x06\n" +
+	"\acompute\x18\x10 \x01(\v2\x11.llmgw.v1.ComputeR\acompute\"\xcc\a\n" +
 	"\x03GPU\x12\x14\n" +
 	"\x05index\x18\x01 \x01(\x05R\x05index\x12\x12\n" +
 	"\x04uuid\x18\x02 \x01(\tR\x04uuid\x12\x12\n" +
@@ -5245,7 +5312,13 @@ const file_llmgw_v1_agent_proto_rawDesc = "" +
 	"\fcluster_uuid\x18\x13 \x01(\tR\vclusterUuid\x12\x1b\n" +
 	"\tclique_id\x18\x14 \x01(\tR\bcliqueId\x12)\n" +
 	"\x10fabric_bandwidth\x18\x15 \x01(\tR\x0ffabricBandwidth\x12%\n" +
-	"\x03mig\x18\x16 \x01(\v2\x13.llmgw.v1.MIGDeviceR\x03mig\x1a;\n" +
+	"\x03mig\x18\x16 \x01(\v2\x13.llmgw.v1.MIGDeviceR\x03mig\x12\x19\n" +
+	"\bpcie_gen\x18\x17 \x01(\x05R\apcieGen\x12 \n" +
+	"\fpcie_gpu_gen\x18\x18 \x01(\x05R\n" +
+	"pcieGpuGen\x12\x1d\n" +
+	"\n" +
+	"pcie_width\x18\x19 \x01(\x05R\tpcieWidth\x12$\n" +
+	"\x0epcie_width_max\x18\x1a \x01(\x05R\fpcieWidthMax\x1a;\n" +
 	"\rNicPathsEntry\x12\x10\n" +
 	"\x03key\x18\x01 \x01(\tR\x03key\x12\x14\n" +
 	"\x05value\x18\x02 \x01(\tR\x05value:\x028\x01\"\x84\x02\n" +
@@ -5361,7 +5434,7 @@ const file_llmgw_v1_agent_proto_rawDesc = "" +
 	"\x06status\x18\x03 \x01(\tR\x06status\x12!\n" +
 	"\fcluster_uuid\x18\x04 \x01(\tR\vclusterUuid\x12\x1b\n" +
 	"\tclique_id\x18\x05 \x01(\tR\bcliqueId\x12\x1c\n" +
-	"\tbandwidth\x18\x06 \x01(\tR\tbandwidth\"\xa9\a\n" +
+	"\tbandwidth\x18\x06 \x01(\tR\tbandwidth\"\xd5\a\n" +
 	"\tHeartbeat\x12\x17\n" +
 	"\anode_id\x18\x01 \x01(\tR\x06nodeId\x12+\n" +
 	"\x12gpu_memory_used_mb\x18\x02 \x03(\x05R\x0fgpuMemoryUsedMb\x12!\n" +
@@ -5383,7 +5456,8 @@ const file_llmgw_v1_agent_proto_rawDesc = "" +
 	"gpu_fabric\x18\x10 \x03(\v2\x18.llmgw.v1.GPUFabricStateR\tgpuFabric\x12\"\n" +
 	"\x04imex\x18\x11 \x01(\v2\x0e.llmgw.v1.IMEXR\x04imex\x12\x1b\n" +
 	"\tcpu_count\x18\x12 \x01(\x05R\bcpuCount\x121\n" +
-	"\x15gpu_foreign_memory_mb\x18\x13 \x03(\x05R\x12gpuForeignMemoryMb\x1aA\n" +
+	"\x15gpu_foreign_memory_mb\x18\x13 \x03(\x05R\x12gpuForeignMemoryMb\x12*\n" +
+	"\x11mem_bandwidth_gbs\x18\x14 \x01(\x01R\x0fmemBandwidthGbs\x1aA\n" +
 	"\x13EngineVersionsEntry\x12\x10\n" +
 	"\x03key\x18\x01 \x01(\tR\x03key\x12\x14\n" +
 	"\x05value\x18\x02 \x01(\tR\x05value:\x028\x01\";\n" +
@@ -5462,7 +5536,7 @@ const file_llmgw_v1_agent_proto_rawDesc = "" +
 	"profile_fp\x18\x0f \x01(\tR\tprofileFp\x12\x17\n" +
 	"\aapi_key\x18\x10 \x01(\tR\x06apiKey\x12\x1f\n" +
 	"\vcpu_threads\x18\x11 \x01(\x05R\n" +
-	"cpuThreads\"\xcd\x03\n" +
+	"cpuThreads\"\xf0\x03\n" +
 	"\rEngineProfile\x12\x12\n" +
 	"\x04name\x18\x01 \x01(\tR\x04name\x12\x16\n" +
 	"\x06engine\x18\x02 \x01(\tR\x06engine\x12\x1b\n" +
@@ -5478,7 +5552,8 @@ const file_llmgw_v1_agent_proto_rawDesc = "" +
 	" \x01(\tR\tmodelPath\x12*\n" +
 	"\x11served_model_name\x18\v \x01(\tR\x0fservedModelName\x121\n" +
 	"\badapters\x18\f \x03(\v2\x15.llmgw.v1.LoraAdapterR\badapters\x12\x1b\n" +
-	"\tkv_events\x18\r \x01(\bR\bkvEvents\x1a6\n" +
+	"\tkv_events\x18\r \x01(\bR\bkvEvents\x12!\n" +
+	"\fallow_pickle\x18\x0e \x01(\bR\vallowPickle\x1a6\n" +
 	"\bEnvEntry\x12\x10\n" +
 	"\x03key\x18\x01 \x01(\tR\x03key\x12\x14\n" +
 	"\x05value\x18\x02 \x01(\tR\x05value:\x028\x01\"%\n" +

api/proto/llmgw/v1/agent.proto

@@ -238,6 +238,9 @@ message Host {
   // nvidia_driver is the driver version, empty when there is none. A card
   // without a driver is not a GPU anything can use.
   string nvidia_driver = 6;
+  // mem_bandwidth_gbs is how fast the memory reads, in GB/s, measured when
+  // the agent started: what decides a model's speed on the CPU. 0 unknown.
+  double mem_bandwidth_gbs = 7;
 }
 
 // Storage is one filesystem a node keeps weights on.
@@ -334,6 +337,13 @@ message GPU {
   // memory the instance's share; utilization_percent stays zero, because no
   // driver measures it per instance.
   MIGDevice mig = 22;
+  // How the card is linked to its host: the fastest generation card and
+  // slot allow together, the card's own fastest, and the lanes in use and
+  // the card's most. Zero when the driver did not say.
+  int32 pcie_gen = 23;
+  int32 pcie_gpu_gen = 24;
+  int32 pcie_width = 25;
+  int32 pcie_width_max = 26;
 }
 
 // MIGDevice is which card a MIG instance is carved from, and how.
@@ -588,6 +598,11 @@ message Heartbeat {
   // is not handed out. Empty from an agent that does not measure it, or from
   // one that could not this time.
   repeated int32 gpu_foreign_memory_mb = 19;
+
+  // mem_bandwidth_gbs is how fast the node's memory reads, in GB/s, once
+  // the agent has measured it — in the background after it starts, so a
+  // registration usually has none yet. 0 means not measured.
+  double mem_bandwidth_gbs = 20;
 }
 
 // Provision is the software a node should have: which engines, at which
@@ -749,6 +764,9 @@ message EngineProfile {
   // kv_events has a vLLM engine publish its KV cache events to the agent, for
   // the router's precise index.
   bool kv_events = 13;
+  // allow_pickle lets the node fetch weights in a pickle format, which run
+  // code when loaded: the catalog's allow_pickle, set on purpose.
+  bool allow_pickle = 14;
 }
 
 // RemoveWeights deletes cached model files a node no longer needs.

docs/admin-api.openapi.json

@@ -2719,6 +2719,57 @@
         }
       }
     },
+    "/api/v1/models/{name}/context": {
+      "post": {
+        "operationId": "postModelsNameContext",
+        "summary": "Change a served model's context: the catalog, its engines' arguments and memory, then its replicas restarted",
+        "tags": [
+          "models"
+        ],
+        "parameters": [
+          {
+            "name": "name",
+            "in": "path",
+            "required": true,
+            "description": "The model's name in the catalog.",
+            "schema": {
+              "type": "string"
+            }
+          },
+          {
+            "$ref": "#/components/parameters/Actor"
+          }
+        ],
+        "requestBody": {
+          "required": true,
+          "content": {
+            "application/json": {
+              "schema": {
+                "$ref": "#/components/schemas/modelserve.ContextRequest"
+              }
+            }
+          }
+        },
+        "responses": {
+          "200": {
+            "description": "OK",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/modelserve.ContextChange"
+                }
+              }
+            }
+          },
+          "401": {
+            "$ref": "#/components/responses/Unauthorized"
+          },
+          "default": {
+            "$ref": "#/components/responses/Error"
+          }
+        }
+      }
+    },
     "/api/v1/models/{name}/eval": {
       "post": {
         "operationId": "postModelsNameEval",
@@ -2913,6 +2964,58 @@
         }
       }
     },
+    "/api/v1/models/{name}/gpus": {
+      "post": {
+        "operationId": "postModelsNameGpus",
+        "summary": "Spread a served model over more cards or fewer: its profile, checked against its KV heads, its cards' memory and the nodes, then its replicas restarted",
+        "description": "Tensor parallel for vLLM, SGLang and TensorRT-LLM, a layer split for llama.cpp, on one node. With no room beside the running replicas, in_place stops each first.",
+        "tags": [
+          "models"
+        ],
+        "parameters": [
+          {
+            "name": "name",
+            "in": "path",
+            "required": true,
+            "description": "The model's name in the catalog.",
+            "schema": {
+              "type": "string"
+            }
+          },
+          {
+            "$ref": "#/components/parameters/Actor"
+          }
+        ],
+        "requestBody": {
+          "required": true,
+          "content": {
+            "application/json": {
+              "schema": {
+                "$ref": "#/components/schemas/modelserve.GPUsRequest"
+              }
+            }
+          }
+        },
+        "responses": {
+          "200": {
+            "description": "OK",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/modelserve.GPUsChange"
+                }
+              }
+            }
+          },
+          "401": {
+            "$ref": "#/components/responses/Unauthorized"
+          },
+          "default": {
+            "$ref": "#/components/responses/Error"
+          }
+        }
+      }
+    },
     "/api/v1/models/{name}/headroom": {
       "get": {
         "operationId": "getModelsNameHeadroom",
@@ -7550,6 +7653,13 @@
           "rate_per_minute": {
             "type": "integer",
             "description": "RatePerMinute bounds each caller's calls to its tools; a negative number takes the bound away."
+          },
+          "clear": {
+            "type": "array",
+            "items": {
+              "type": "string"
+            },
+            "description": "Clear empties what an empty field above would leave as it is: \"description\", \"timeout\" (back to the default), \"rate\"."
           }
         }
       },
@@ -7742,6 +7852,13 @@
           "name": {
             "type": "string"
           },
+          "context_length": {
+            "type": "integer",
+            "description": "ContextLength is what one request may use, MaxOutput what it may answer; 0 when the catalog does not say."
+          },
+          "max_output": {
+            "type": "integer"
+          },
           "canary": {
             "$ref": "#/components/schemas/catalog.Canary",
             "description": "Canary is a share of its requests sent to a candidate; CanaryOf is set on the candidate, naming the model it is a canary of."
@@ -8011,6 +8128,10 @@
             "type": "integer",
             "description": "CPUCount is the machine's CPUs, 0 when the agent did not report them."
           },
+          "mem_bandwidth_gbs": {
+            "type": "number",
+            "description": "MemBandwidthGBs is how fast its memory reads, in GB/s, measured when its agent started; 0 when not reported. A model's speed on the CPU."
+          },
           "engines": {
             "type": "array",
             "items": {
@@ -9930,6 +10051,10 @@
             "type": "string",
             "description": "Source is the weights path or a HuggingFace id."
           },
+          "allow_pickle": {
+            "type": "boolean",
+            "description": "AllowPickle lets the nodes fetch this model's weights, or its adapters', in a pickle format (.bin, .pt, .pth, .ckpt), which run code as they are loaded. Off, such a source is refused when the model is added and by the node before it fetches anything. For a model that has no safetensors or GGUF form and whose files were checked."
+          },
           "profiles": {
             "type": "array",
             "items": {
@@ -10309,6 +10434,10 @@
             "type": "string",
             "description": "HFRepo is the repository its node loads the weights by name from."
           },
+          "allow_pickle": {
+            "type": "boolean",
+            "description": "AllowPickle is the member model's allow_pickle."
+          },
           "workflow": {
             "type": "string",
             "description": "Workflow and Checkpoint are what a request for it is sent."
@@ -10986,6 +11115,137 @@
           }
         }
       },
+      "modelserve.ContextChange": {
+        "type": "object",
+        "description": "ContextChange is what was changed, and the replicas started again on it.",
+        "properties": {
+          "model": {
+            "type": "string"
+          },
+          "before": {
+            "type": "integer"
+          },
+          "after": {
+            "type": "integer"
+          },
+          "max_output": {
+            "type": "integer"
+          },
+          "trained": {
+            "type": "integer"
+          },
+          "profiles": {
+            "type": "array",
+            "items": {
+              "$ref": "#/components/schemas/modelserve.ContextProfile"
+            }
+          },
+          "restarted": {
+            "type": "integer"
+          },
+          "note": {
+            "type": "string",
+            "description": "Note is what could not be checked, said once."
+          }
+        }
+      },
+      "modelserve.ContextProfile": {
+        "type": "object",
+        "description": "ContextProfile is what one profile needs before and after.",
+        "properties": {
+          "profile": {
+            "type": "string"
+          },
+          "engine": {
+            "type": "string"
+          },
+          "before": {
+            "type": "string"
+          },
+          "after": {
+            "type": "string"
+          },
+          "args": {
+            "type": "array",
+            "items": {
+              "type": "string"
+            }
+          }
+        }
+      },
+      "modelserve.ContextRequest": {
+        "type": "object",
+        "description": "ContextRequest is the context to change to: tokens one request may use, and for llama.cpp how many at once (its pool is the two multiplied).",
+        "properties": {
+          "tokens": {
+            "type": "integer"
+          },
+          "slots": {
+            "type": "integer"
+          }
+        }
+      },
+      "modelserve.GPUsChange": {
+        "type": "object",
+        "description": "GPUsChange is what was changed, and the replicas started again on it.",
+        "properties": {
+          "model": {
+            "type": "string"
+          },
+          "profile": {
+            "type": "string"
+          },
+          "engine": {
+            "type": "string"
+          },
+          "gpu_class": {
+            "type": "string"
+          },
+          "before": {
+            "type": "integer"
+          },
+          "after": {
+            "type": "integer"
+          },
+          "args": {
+            "type": "array",
+            "items": {
+              "type": "string"
+            }
+          },
+          "per_card": {
+            "type": "string",
+            "description": "PerCard is what each card holds after, when it could be worked out."
+          },
+          "restarted": {
+            "type": "integer"
+          },
+          "in_place": {
+            "type": "boolean"
+          },
+          "note": {
+            "type": "string",
+            "description": "Note is what could not be checked, said once."
+          }
+        }
+      },
+      "modelserve.GPUsRequest": {
+        "type": "object",
+        "description": "GPUsRequest is how many cards a replica is to use.",
+        "properties": {
+          "gpus": {
+            "type": "integer"
+          },
+          "profile": {
+            "type": "string",
+            "description": "Profile is the one to change, when the model has more than one profile on cards."
+          },
+          "in_place": {
+            "type": "boolean",
+            "description": "InPlace stops each replica before its replacement starts: for when no node has the cards free beside it. It does not answer while the new one loads."
+          }
+        }
+      },
       "modelserve.Job": {
         "type": "object",
         "description": "Job is one model being served, and where it stands.",

docs/console.md

@@ -173,6 +173,23 @@ search…** as an administrator, or **Delete…** it. People
 fill collections from their own machines with `llmgw-kb`
 ([knowledge.md](knowledge.md)). Off on a SQLite store, and says why.
 
+**MCP servers** lists the company's own MCP servers that agents reach through
+the gateway's `/mcp`, their tools named `server.tool`: whether each is on,
+who has been granted its tools, its rate per caller, the names of the
+credentials it is sent (never their values) and its address. **A** adds one:
+its name, which is its tools' prefix, its address, and a bearer token or
+another header, kept sealed in the store. On a row: **Test, and list its
+tools**, with its credential, as a call would; **Grant its tools to…** a
+team, project, organization, person or key — all of them, `jira.*`, or some
+by name — beside whatever else that level is granted, and saying when a
+level above keeps them out of reach; **Revoke** a grant;
+**Change…** its address, timeout or rate; **Replace its credential…**;
+**Turn off** (its tools are listed nowhere and every call refused) or on;
+**Remove…**. A server's tools are seen nowhere until granted. The screen
+also says how a person points an agent at the gateway, with a key of their
+own from `/profile/keys`. `llmgw ctl mcp` does the same from a shell, and
+the web page **MCP servers** (`/admin/mcp`, administrators only) in a browser.
+
 **Mirror** is what the model mirror holds, one row for each version of each
 model — a file, or the set of files a model of several is: shards, a vision
 projector, a config and tokenizer — with how far its download has got and

docs/guardrails.md

@@ -301,10 +301,18 @@ policies:
 string; empty sends `{"input": …}`). `blocked` and `reason` are dotted paths
 into the JSON answer, numbers for list items; true, a number above zero, a
 non-empty list or a string other than "false" is unsafe. On input the
-service reads the latest user turn; on output the whole answer, once, at its
-end — a streamed answer released a sentence at a time is then stopped or
-flagged as the judge's is. The rule recorded is `service:<name>:<reason>`.
-The service is told the text only, never who sent it.
+service reads the latest user turn as the model will, with what `redact`
+took out still out; on output the whole answer, once, at its end, as the
+client reads it — with `restore`, the values put back. Being at the end, a
+service can stop a streamed answer only where it is held back
+(`stream: buffer_all`): released a sentence at a time, the client has read
+it by then, and what the service finds ends the stream with
+`content_filter`, flags it and is recorded. The rule recorded is
+`service:<name>:<reason>`. The service is told the text only, never who
+sent it. Under `fail: open`, the default, a service that cannot be asked
+lets the text through — and a caller can make it fail, sending what it
+refuses to read; give `fail: closed` where the service must decide. Of an
+answer asked for with `n` above 1, the service reads the first.
 
 ## Where a policy applies
 

docs/install.md

@@ -876,6 +876,27 @@ Nothing to do here if you built the mirror in step 3 — the agent fetches what
 model's `source` names, and that source is your mirror. This section is for the
 arrangements where the files are already on the machine.
 
+**Weights in a pickle format are refused.** `.bin`, `.pt`, `.pth`, `.ckpt` and
+`.pkl` are Python pickles, and loading one runs whatever it holds, as the
+engine's user, on the node — which is how the malicious models found on Hugging
+Face work. A model whose source is one, or a mirror version holding one, is
+refused when it is added and by the node before it fetches anything; vLLM and
+SGLang given a model by its Hugging Face id read only its safetensors. Take the
+repository's safetensors or GGUF. A model that has no other form, and whose
+files were checked, can say so:
+
+```yaml
+- name: triposr
+  source: http://mirror:8081/triposr/<rev>/model.ckpt
+  allow_pickle: true
+```
+
+A local path is judged by its name as any source is; a local directory is
+not looked into. What an engine downloads itself by a Hugging Face id, past
+the node — an adapter given by its id, a TensorRT-LLM model's — is refused
+when the model is added, unless it says `allow_pickle`: mirror its
+safetensors and give that instead.
+
 Put them where the node can read them. `scp`, a shared filesystem, or the
 engine's own downloader are all valid, and which you want depends on how many
 nodes you have.
@@ -1034,6 +1055,27 @@ A node with no location is placed as before, as a rack of its own.
 See [facility-aware-placement.md](facility-aware-placement.md) for power caps
 and holds.
 
+### A card in the wrong slot
+
+An x16 card in a slot wired x8 or x4, behind a riser, or in a slot of an
+older PCIe generation than the card's, links slower than it was built to and
+works otherwise. A model on the card still decodes at the card's own memory
+speed, but it loads slower, a replica split over several cards by PCIe is
+slower, and so is anything that spills to the host. Each card's agent reads
+how its link stands — the fastest generation card and slot allow together,
+and the lanes in use — against what the card could do, and the node says so
+among its warnings (the dashboard, the console's Nodes, `llmgw doctor`):
+
+```
+GPU 1 (NVIDIA H100 80GB HBM3) linked at x8 of its x16 lanes: 31.5 GB/s where it
+could have 63.0. Check the slot (wired fewer lanes, or an older generation), a
+riser, or the BIOS's PCIe setting
+```
+
+The generation a card reports now drops while it idles, to save power; what is
+compared is the fastest it may reach, which does not. `lspci -vv -s BUS-ID`
+shows the same: `LnkCap` is the card's, `LnkSta` the link's.
+
 ## 6. Setting the node's CPU model, if it is a VM
 
 Proxmox defaults a VM to `kvm64`, which hides AVX, AVX2, FMA and even SSE4.2
@@ -1587,6 +1629,55 @@ Not yet: models declared in server.yaml cannot name a peer's model as a
 fallback (the catalog in the file is checked before the peers are read), and
 a peer behind a private CA needs that CA in the system's trust store.
 
+### Changing a served model's context
+
+How much one request may use is set when a model is served. To change it
+afterwards:
+
+```
+llmgw ctl model context qwen-7b 32768             # tokens a request
+llmgw ctl model context qwen-7b 32768 --slots 4   # llama.cpp: four at once
+```
+
+The catalog, the engine's arguments (`--ctx-size` and `--parallel` for
+llama.cpp, `--max-model-len` for vLLM, `--context-length` for SGLang) and
+the memory its KV cache takes change together. A context longer than the
+model was trained for is refused, and so is one its cards cannot hold — with
+the figures, so you know how far to go. Its replicas are then started again on
+it, one by one, a new one ready before the old one stops. The console has the
+same under a model's menu, Context length…. A model declared in the cluster's
+settings is changed there (`llmgw ctl config edit`).
+
+### Moving a served model to more cards
+
+How many cards of one node each replica uses can be changed while the model
+is served:
+
+```
+llmgw ctl model gpus qwen3-coder-30b 2              # two cards a replica
+llmgw ctl model gpus llama-70b 4 --profile h100     # the model has several profiles on cards
+llmgw ctl model gpus qwen3-coder-30b 2 --in-place   # no room beside the running replica
+```
+
+vLLM, SGLang and TensorRT-LLM spread the model by tensor parallel (`tp`),
+which needs the count to divide the model's KV heads or be divided by them:
+a model with 8 splits 2, 4, 8 or 16 ways, not 3 or 12. llama.cpp has no
+tensor parallel; it splits the layers over the cards (`gpus_per_node`), and
+a `--tensor-split` written for the old cards is dropped. ComfyUI and Strata
+run on one card. A count no node has is refused, and so is one whose cards
+would not hold the weights and the context — with the figures; fewer cards
+with a long context may need `llmgw ctl model context` first.
+
+Its replicas are then started again on it, each new one ready beside the old
+before the old stops. That needs the cards free beside it: on a node of two
+cards with the model on one, two do not fit until it is gone, and the command
+says so. `--in-place` stops each replica first and starts its replacement on
+the cards it held and the free ones, and the model does not answer while the
+new one loads — a model with several replicas keeps answering on the others.
+A replica spread over several nodes is laid out by hand, with a fabric
+(`nodes`, `gpus_per_node` in the profile). The console has the same under a
+model's menu, GPUs….
+
 ### Trying new weights on some of the traffic first
 
 When a model's repository publishes new weights, `llmgw ctl model update`
@@ -2137,8 +2228,10 @@ llmgw keys cache --project Acme/ml/train --off
 - **Policies still apply:** the limits are checked and the guardrails judge
   the request before the cache is looked in, and the output policy judges a
   reused answer as it would a new one. A request continuing a stored
-  conversation, or under a `no_store` policy, is never answered from the
-  cache nor kept in it.
+  conversation, under a `no_store` policy, or whose redacted values are put
+  back in its answer (`restore: true`) is never answered from the cache nor
+  kept in it: redacted, two callers' requests read the same, and the answer
+  has the first caller's values in it.
 - **What a hit costs:** nothing. It is recorded as a request — status 200,
   the model, the key, the rate limit's count — with no tokens and no cost,
   marked as a cache hit (`llmgw ctl request <id>` says so), so it spends no
@@ -2178,8 +2271,8 @@ router:
   `router.cache.embedding_model`, against the questions already answered
   in the same conversation — the same system prompt, the same earlier turns,
   tools, `response_format` and sampling settings, the same model and
-  project. A like question in another conversation, or after a tool's
-  result, is not taken for this one. The exact cache is asked first, and an
+  project. A like question in another conversation, after a tool's result,
+  or with a picture, a sound or a file in it, is not taken for this one. The exact cache is asked first, and an
   exact hit embeds nothing.
 - **How alike:** the cosine similarity of the two embeddings, at least the
   project's `--similar`. 0.95 is strict; below 0.9 different questions start
@@ -2306,6 +2399,7 @@ triposr`. What that adds is this entry, and its workflow is built in:
 ```yaml
 - name: triposr
   source: hf:stabilityai/TripoSR/model.ckpt
+  allow_pickle: true               # published as this checkpoint only
   comfyui:
     workflow: triposr              # built in; a model of your own names a JSON file on every control plane
     output: meshes
@@ -2477,8 +2571,31 @@ lost: a standby leads within seconds, Patroni promotes the standby database
 if it was the primary, and the floating address follows whichever leads. The
 mirror's management API answers the three control planes' addresses, not
 only its own machine's, so a standby that comes to lead can still fetch and
-add models. The mirror itself stays on the first machine: while it is down,
-models the nodes already have keep serving, and new ones wait for it.
+add models.
+
+**A second mirror.** The mirror is on the first machine, and without a
+second one, models the nodes already have keep serving while that machine
+is down, and new ones wait for it. With a second, the cluster carries on.
+- **Setting it up:** the screens offer one on either database machine
+  ("A second mirror"); by command, `--second-mirror 1` (or `2`). For an
+  install that has one mirror, run this on the mirror's machine:
+
+  ```
+  sudo llmgw mirror add-peer ubuntu@10.0.0.7 --control-plane 10.0.0.5 --control-plane 10.0.0.6 --control-plane 10.0.0.7
+  ```
+- **Kept in step:** each mirror reads the other every thirty seconds and
+  copies what it lacks, keeping a copy only if its checksum matches.
+  - A model removed from one is removed from the other, even if that one
+    was down at the time and comes back later.
+  - A model added again after it was removed stands on both.
+  - Neither mirror leads.
+- **Control planes:** they name both (`mirror.peers` in the cluster's
+  settings) and send the mirror's management calls to the first that
+  answers.
+- **Nodes:** while the mirror a model's address names is down, a node is
+  given the same file on the other.
+- **Disk:** the second machine needs room for every model, as the first
+  does.
 
 ## Two control planes
 
@@ -2586,10 +2703,20 @@ llmgw keys limits --scope team --id Acme/support --tools 'jira.*'
 - **Who may call what:** tools are named `server.tool` and granted as
   patterns (`jira.*`, `db.query`) at any level of the tenancy tree, as
   models are.
-  - A level that names any narrows what is above it.
+  - A level that names any narrows what is above it: a tool is seen where
+    every level above that names tools names it too. A team under an
+    organization that names `hr.*` sees no `jira.*` whatever the team is
+    given, until the organization names it as well.
   - Unlike models, **none granted anywhere is none**: a tool reaches into a
     company system, so it is granted, never assumed.
-  - The console's limits form has the same field.
+  - `keys limits --tools` writes a level's list whole, every server's
+    together; an empty list (`--tools ''`) closes the level to every tool.
+    The console's **MCP servers** screen and the `/admin/mcp` page grant
+    and take back one server's tools: a level given its first list starts
+    from the nearest list above it, so it keeps what it saw; one left
+    naming none goes back to naming none, or stays closed when a level
+    above names tools, so taking one server away opens it to no other. They
+    say when a level above keeps the granted tools out of reach.
 - **The server is told who calls:** `X-Llmgw-Key`, and `X-Llmgw-User` and
   `X-Llmgw-Project` where there are some. It is called with llmgw's
   credentials, never the caller's key.

docs/quick-install.md

@@ -146,7 +146,35 @@ agent is stopped.
 ## 4. Add a model
 
 Open `/admin/mirror` and choose **Explore models**. It lists what is popular on
-Hugging Face, showing only what your nodes can run.
+Hugging Face, showing only what your nodes can run: judged by the cluster as it
+is now — the largest card and its compute capability, the most memory on one
+node — so a model too big for them, in a form those cards lack (FP8 needs 8.9),
+or in neither GGUF nor safetensors is left out, and the page says how many.
+Without GPUs it lists GGUF only. Untick **only what these nodes can run** to
+see everything. On a model's page a file too big for the cluster is shown
+without a **Fetch** button, with what it would need.
+
+The **CPU tok/s** column is about how fast one reply would come on the CPU of
+the fastest node whose memory holds the model at Q4_K_M; a model's page gives
+it for each GGUF file by its own size. Every node's agent measures how fast its
+memory reads when it starts (it is in the agent's log as `memory_gbs`), and a
+token on a CPU reads every weight once: the estimate is that speed over the
+model's size — for a model of experts named like `30B-A3B`, its active part.
+On the test cluster it came within 15% of what llama.cpp measured. Several
+replies at once share the reads and each is slower.
+
+**GPU tok/s** is the same on the cards, for the layout this cluster would
+give the model — the fewest cards that hold it, of each class, the fastest of
+those; its tooltip says which ("2× h100-80g, NVLink, vLLM", "2 nodes × 1×
+a100-80g over ib0 (InfiniBand 200 Gb/s), vLLM"). A card's memory bandwidth is
+known from its model name (an H100 SXM and an H100 PCIe differ by half); a card
+not in the table gets no figure. Tensor parallel across a node's cards divides
+what each reads, and adds two all-reduces a layer, short over NVLink and
+longer over PCIe; llama.cpp across cards splits by layer, so the cards take
+turns and a reply is no faster than on one; a pipeline across nodes takes turns
+too, and adds a hop over the fabric for each node. One card's figures were
+measured on the test cluster's RTX 6000; tensor parallel and the fabric are
+from what NCCL is known to take, not measured here.
 
 1. Click a model to read its card.
 2. Pick one of its files, usually a `q4_k_m` for a first try, and click

docs/reference.md

@@ -180,6 +180,7 @@ rather than repeated.
 | `parallel` | `int` | [LlamaCppConfig](#llamacppconfig) |
 | `threads` | `int` | [LlamaCppConfig](#llamacppconfig) |
 | `api` | `string` | [Mirror](#mirror) |
+| `peers` | `[]string` | [Mirror](#mirror) |
 | `tls` | `tlsconf.Config` | [Mirror](#mirror) |
 | `registry` | `string` | [Mirror](#mirror) |
 | `timeout` | `time.Duration` | [Mirror](#mirror) |
@@ -188,6 +189,7 @@ rather than repeated.
 | `aliases` | `[]string` | [Model](#model) |
 | `description` | `string` | [Model](#model) |
 | `source` | `string` | [Model](#model) |
+| `allow_pickle` | `bool` | [Model](#model) |
 | `profiles` | `[]Profile` | [Model](#model) |
 | `scaling` | `Scaling` | [Model](#model) |
 | `class` | `string` | [Model](#model) |
@@ -394,6 +396,9 @@ rather than repeated.
 | `timeout` | `time.Duration` | [Service](#service) |
 | `stopped` | `bool` | [ServingSettings](#servingsettings) |
 | `bin` | `string` | [SimConfig](#simconfig) |
+| `dir` | `string` | [StrataConfig](#strataconfig) |
+| `python` | `string` | [StrataConfig](#strataconfig) |
+| `engine` | `string` | [StrataConfig](#strataconfig) |
 | `python` | `string` | [TRTLLMConfig](#trtllmconfig) |
 | `store` | `string` | [Threads](#threads) |
 | `retain_hours` | `int` | [Threads](#threads) |
@@ -474,6 +479,7 @@ rather than repeated.
 | `trtllm` | `TRTLLMConfig` | [engines.Config](#enginesconfig) |
 | `sim` | `SimConfig` | [engines.Config](#enginesconfig) |
 | `comfyui` | `ComfyUIConfig` | [engines.Config](#enginesconfig) |
+| `strata` | `StrataConfig` | [engines.Config](#enginesconfig) |
 | `enabled` | `*bool` | [guard.Config](#guardconfig) |
 | `default` | `[]string` | [guard.Config](#guardconfig) |
 | `policies` | `map[string]Policy` | [guard.Config](#guardconfig) |
@@ -1721,6 +1727,14 @@ The management port, not the one nodes fetch bytes from. Those are two
 listeners on the mirror precisely so this one can be locked down without
 re-certifying every agent's fetcher.
 
+### `peers`
+
+`[]string`
+
+Peers are the management addresses of mirrors kept in step with this
+one (llmgw mirror serve --peer): asked when it cannot be reached, and
+a node given the same file on one of them while it is down.
+
 ### `tls`
 
 `tlsconf.Config`
@@ -1790,6 +1804,16 @@ choice by guesswork.
 
 Source is the weights path or a HuggingFace id.
 
+### `allow_pickle`
+
+`bool`
+
+AllowPickle lets the nodes fetch this model's weights, or its
+adapters', in a pickle format (.bin, .pt, .pth, .ckpt), which run code
+as they are loaded. Off, such a source is refused when the model is
+added and by the node before it fetches anything. For a model that has
+no safetensors or GGUF form and whose files were checked.
+
 ### `profiles`
 
 `[]Profile`
@@ -3395,6 +3419,27 @@ entries ask. In force on every control plane within seconds.
 Bin is the llmgw binary itself; the engine is one of its subcommands.
 Empty means whatever started this agent, which is right almost always.
 
+## StrataConfig
+
+### `dir`
+
+`string`
+
+Dir is Strata's source directory, where setup.py and serve/ are.
+
+### `python`
+
+`string`
+
+Python is the interpreter of its environment; Dir's .venv when unset.
+
+### `engine`
+
+`string`
+
+Engine is the directory of its prebuilt engine, where the strata
+binary and the CUDA libraries it ships are; Dir's engine/ when unset.
+
 ## TRTLLMConfig
 
 ### `python`
@@ -3967,6 +4012,15 @@ ComfyUI makes pictures; offered by a node whose ComfyUI is named here.
 
 Takes a [ComfyUIConfig](#comfyuiconfig) block.
 
+### `strata`
+
+`StrataConfig`
+
+Strata serves a few mixture-of-experts models from RAM and one card;
+offered by a node whose Strata is named here, or that installed it.
+
+Takes a [StrataConfig](#strataconfig) block.
+
 ## guard.Config
 
 ### `enabled`

internal/agent/agent.go

@@ -12,6 +12,7 @@ import (
 	"io"
 	"log/slog"
 	"maps"
+	"math"
 	"math/rand/v2"
 	"net"
 	"net/http"
@@ -392,8 +393,11 @@ func (a *Agent) settleAdopted(sh *supervisor.Shard, o supervisor.Orphan) {
 		_ = a.sup.Stop(ctx, o.ShardID)
 		return
 	}
-	// What it computes on, for one started before the agent kept that.
-	if a.sup.ComputeOf(o.ShardID) == nil {
+	// What it computes on, for one started before the agent kept that — or
+	// kept as the CPU for a GPU placement by an agent that did not follow
+	// an engine into a process group of its own, which the check above has
+	// just found on its GPUs.
+	if c := a.sup.ComputeOf(o.ShardID); c == nil || (len(o.Spec.GPUIndices) > 0 && c.Backend == "CPU") {
 		if c, ok := a.gpuCompute(ctx, o.Spec.GPUIndices, sh.Status().PID); ok {
 			a.sup.SetCompute(o.ShardID, c)
 		}
@@ -513,6 +517,7 @@ func (a *Agent) scan(ctx context.Context) error {
 	inv, err := discovery.Discover(ctx, opts, a.log)
 	host := discovery.DiscoverHost()
 	host.MinComputeCapability = discovery.MinComputeCapability(inv.GPUs)
+	host.MemBandwidthGBs = a.memBandwidth()
 	a.mu.Lock()
 	a.inv = inv
 	a.prov.Host = host
@@ -569,6 +574,9 @@ func (a *Agent) session(ctx context.Context, server string) error {
 	reg.Host = &llmgwv1.Host{
 		OsId: h.OSID, OsVersion: h.OSVersion, OsCodename: h.OSCodename,
 		Arch: h.Arch, CpuFlags: h.CPUFlags, NvidiaDriver: h.NVIDIADriver,
+		// What is measured by now, not what the startup scan found before
+		// the measurement ended: a reconnect would otherwise say 0.
+		MemBandwidthGbs: a.memBandwidth(),
 	}
 
 	if err := stream.Send(&llmgwv1.AgentMessage{
@@ -666,6 +674,7 @@ func (a *Agent) heartbeat() *llmgwv1.AgentMessage {
 	// without anybody restarting anything.
 	hb.MemoryTotalMb = int64(discovery.MemoryTotalMB())
 	hb.CpuCount = int32(runtime.NumCPU())
+	hb.MemBandwidthGbs = a.memBandwidth()
 
 	a.reprobeEngines()
 	a.mu.Lock()
@@ -1124,7 +1133,7 @@ func (a *Agent) startShard(ctx context.Context, m *llmgwv1.StartShard, out chan<
 			},
 		})
 	})
-	if path, err := a.weights().Ensure(ctx, spec.Profile.ModelPath); err != nil {
+	if path, err := a.weights().Ensure(ctx, spec.Profile.ModelPath, spec.Profile.AllowPickle); err != nil {
 		stop()
 		report(supervisor.Failed, err.Error(), nil)
 		return
@@ -1145,7 +1154,7 @@ func (a *Agent) startShard(ctx context.Context, m *llmgwv1.StartShard, out chan<
 				send(&llmgwv1.ShardStatus{State: agentsvc.StateToWire(supervisor.Starting),
 					Fetch: &llmgwv1.FetchProgress{Source: p.Source, BytesDone: p.BytesDone, BytesTotal: p.BytesTotal}})
 			})
-			path, err := a.weights().Ensure(ctx, m["source"])
+			path, err := a.weights().Ensure(ctx, m["source"], m["allow_pickle"] == "true")
 			stop()
 			if err != nil {
 				report(supervisor.Failed, fmt.Sprintf("model %s: %v", m["name"], err), nil)
@@ -1161,7 +1170,7 @@ func (a *Agent) startShard(ctx context.Context, m *llmgwv1.StartShard, out chan<
 	// loads them at start, and one missing is a replica that answers for the
 	// base model under the adapter's name.
 	for i, ad := range spec.Profile.Adapters {
-		path, err := a.weights().Ensure(ctx, ad.Path)
+		path, err := a.weights().Ensure(ctx, ad.Path, spec.Profile.AllowPickle)
 		if err != nil {
 			report(supervisor.Failed, fmt.Sprintf("adapter %s: %v", ad.Name, err), nil)
 			return
@@ -1505,18 +1514,25 @@ func ResolveManagement(c Config) (addr, iface string) {
 // shard computes on as soon as that is known: an adopted engine's is found
 // after registration, which carried what was known then.
 func (a *Agent) reportCompute(ctx context.Context, out chan<- *llmgwv1.AgentMessage) {
-	sent := map[string]bool{}
+	// What was sent for each shard, so a compute that changes is sent again:
+	// an adopted shard's kept one is sent at once, and adopting it may then
+	// find it was kept wrong.
+	sent := map[string]string{}
 	t := time.NewTicker(5 * time.Second)
 	defer t.Stop()
 	for range 60 {
 		for _, st := range a.sup.Statuses() {
-			if st.State.Terminal() || sent[st.ID] {
+			if st.State.Terminal() {
 				continue
 			}
 			c := a.sup.ComputeOf(st.ID)
 			if c == nil {
 				continue
 			}
+			said := fmt.Sprint(*c)
+			if sent[st.ID] == said {
+				continue
+			}
 			sh, err := a.sup.Get(st.ID)
 			if err != nil {
 				continue
@@ -1527,7 +1543,7 @@ func (a *Agent) reportCompute(ctx context.Context, out chan<- *llmgwv1.AgentMess
 			}}}
 			select {
 			case out <- msg:
-				sent[st.ID] = true
+				sent[st.ID] = said
 			case <-ctx.Done():
 				return
 			}
@@ -1539,3 +1555,38 @@ func (a *Agent) reportCompute(ctx context.Context, out chan<- *llmgwv1.AgentMess
 		}
 	}
 }
+
+// memBandwidth is the node's memory read rate in GB/s, 0 until measured.
+// Measured in the background, the first time it is asked for: it reads half
+// a gigabyte, which must not hold up an agent reaching its control plane,
+// and the heartbeat carries it when it is known. Not at once but after a
+// minute or so, chosen at random, and again every hour, keeping the best:
+// VMs on one host that all restart together measure each other, and read a
+// third less than any of them gets alone. Once a process rather than an
+// agent, so tests running many agents in one read it once.
+func (a *Agent) memBandwidth() float64 {
+	bwOnce.Do(func() {
+		go func() {
+			time.Sleep(bandwidthDelay())
+			for {
+				if bw := measureBandwidth(); bw > math.Float64frombits(bwBits.Load()) {
+					bwBits.Store(math.Float64bits(bw))
+					a.log.Info("this node's memory", "read_gbs", math.Round(bw*10)/10)
+				}
+				time.Sleep(time.Hour)
+			}
+		}()
+	})
+	return math.Float64frombits(bwBits.Load())
+}
+
+// bandwidthDelay is how long an agent waits before measuring, so agents
+// started together measure apart.
+var bandwidthDelay = func() time.Duration { return 10*time.Second + rand.N(110*time.Second) }
+
+var (
+	bwOnce sync.Once
+	bwBits atomic.Uint64
+	// measureBandwidth is replaced in tests, which have no use for it.
+	measureBandwidth = discovery.MeasureMemoryBandwidth
+)

internal/agent/agent_test.go

@@ -671,7 +671,7 @@ func TestAShardWithMissingWeightsFailsBeforeLaunching(t *testing.T) {
 	// nobody is watching — so it is caught here, where the message can say
 	// what is actually wrong.
 	a := &Agent{cfg: Config{WeightsCache: t.TempDir()}, log: slog.New(slog.DiscardHandler)}
-	_, err := a.weights().Ensure(t.Context(), filepath.Join(t.TempDir(), "not-there.gguf"))
+	_, err := a.weights().Ensure(t.Context(), filepath.Join(t.TempDir(), "not-there.gguf"), false)
 	if err == nil {
 		t.Fatal("a missing weights file was accepted")
 	}

internal/agent/discovery/gpu.go

@@ -52,6 +52,8 @@ type GPU struct {
 	// memory the instance's share. Utilisation stays unreported, because no
 	// driver measures it per instance.
 	MIG *MIGDevice `json:"mig,omitempty"`
+	// PCIe is how the card is linked to its host, zero when not read.
+	PCIe PCIeLink `json:"pcie,omitempty"`
 }
 
 // GPUHealth records conditions that make a GPU unsafe to schedule onto.

internal/agent/discovery/gpu_nvml.go

@@ -67,10 +67,23 @@ func DiscoverGPUsNVML(_ context.Context) ([]GPU, error) {
 		if major, minor, ret := dev.GetCudaComputeCapability(); ret == nvml.SUCCESS {
 			g.ComputeCapability = fmt.Sprintf("%d.%d", major, minor)
 		}
+		if v, ret := dev.GetMaxPcieLinkGeneration(); ret == nvml.SUCCESS {
+			g.PCIe.Gen = v
+		}
+		if v, ret := dev.GetGpuMaxPcieLinkGeneration(); ret == nvml.SUCCESS {
+			g.PCIe.GPUGen = v
+		}
+		if v, ret := dev.GetCurrPcieLinkWidth(); ret == nvml.SUCCESS {
+			g.PCIe.Width = v
+		}
+		if v, ret := dev.GetMaxPcieLinkWidth(); ret == nvml.SUCCESS {
+			g.PCIe.WidthMax = v
+		}
 		if pci, ret := dev.GetPciInfo(); ret == nvml.SUCCESS {
 			g.PCIBusID = busIDString(pci.BusId[:])
 			byBusID[normalizeBusID(g.PCIBusID)] = i
 		}
+		g.PCIe = withSysfs(g.PCIe, "/sys/bus/pci/devices", g.PCIBusID)
 		if t, ret := dev.GetTemperature(nvml.TEMPERATURE_GPU); ret == nvml.SUCCESS {
 			g.Health.TemperatureC = int(t)
 		}

internal/agent/discovery/gpu_smi.go

@@ -64,6 +64,7 @@ func DiscoverGPUsSMI(ctx context.Context, binary string) ([]GPU, error) {
 	if err != nil || len(cards) == 0 {
 		return cards, err
 	}
+	readPCIe(ctx, binary, cards)
 	// The query lists whole cards, MIG mode or not. A card in MIG mode is
 	// replaced by its instances, which only -L names and only the full
 	// report sizes; asked only when some card could be in MIG mode, since

internal/agent/discovery/host.go

@@ -31,6 +31,10 @@ type Host struct {
 	// MinComputeCapability is the oldest GPU's, 0 when there are none or it
 	// is not known. Not read here but from the GPU inventory, by the agent.
 	MinComputeCapability float64 `json:"min_compute_capability,omitempty"`
+	// MemBandwidthGBs is how fast the memory reads, in GB/s, 0 when not
+	// measured: what decides a model's speed on the CPU. Not read here but
+	// measured by the agent when it starts (MeasureMemoryBandwidth).
+	MemBandwidthGBs float64 `json:"mem_bandwidth_gbs,omitempty"`
 }
 
 // Family is the packaging family: "deb", "rpm", or "" when neither is known.

internal/agent/discovery/membw.go

@@ -0,0 +1,146 @@
+package discovery
+
+import (
+	"bufio"
+	"os"
+	"path/filepath"
+	"runtime"
+	"strconv"
+	"strings"
+	"sync"
+	"time"
+)
+
+// How fast this machine reads its memory, which on a CPU is what decides how
+// fast a model answers: every token of a reply reads all the weights it uses
+// once, so a node that reads 40 GB/s writes a 4 GB model's tokens about ten
+// times a second, whatever its core count. Measured once when the agent
+// starts, in a fraction of a second, by every CPU summing a share of a buffer
+// too large for any cache — a VM sharing its host's memory with others gets
+// its share, which is the number that matters for it.
+
+// bandwidthBuffer is what is read at most; the least is eight times the
+// processor's last cache, and never under minBuffer. Smaller, it would be
+// the cache that was measured — some hold a gigabyte — and the figure kept
+// would be too high.
+const (
+	bandwidthBuffer = 2 << 30
+	minBuffer       = 256 << 20
+)
+
+// MeasureMemoryBandwidth is the memory's read rate in GB/s, the best of a
+// few passes; 0 when the machine has too little free memory to measure.
+func MeasureMemoryBandwidth() float64 {
+	size := int64(bandwidthBuffer)
+	// Never more than a quarter of what is free: a node already holding
+	// models must not be pushed into swap by a measurement. Without room
+	// for a buffer past the cache, none is measured: the hourly retry
+	// measures it when there is.
+	if avail := memAvailable("/proc/meminfo"); avail > 0 {
+		size = min(size, avail/4)
+	}
+	if size < max(minBuffer, 8*lastCacheBytes("/sys/devices/system/cpu")) {
+		return 0
+	}
+	buf := make([]uint64, size/8)
+	for i := range buf {
+		buf[i] = uint64(i)
+	}
+	workers := runtime.NumCPU()
+	chunk := (len(buf) + workers - 1) / workers
+	best := 0.0
+	for pass := 0; pass < 4; pass++ {
+		var wg sync.WaitGroup
+		start := time.Now()
+		for w := 0; w < workers; w++ {
+			lo, hi := w*chunk, min((w+1)*chunk, len(buf))
+			if lo >= hi {
+				continue
+			}
+			wg.Add(1)
+			go func(part []uint64) {
+				defer wg.Done()
+				sink(sum(part))
+			}(buf[lo:hi])
+		}
+		wg.Wait()
+		if s := time.Since(start).Seconds(); s > 0 {
+			best = max(best, float64(size)/s/1e9)
+		}
+	}
+	return best
+}
+
+// sum reads every word, eight at a time so the loop is not what limits it.
+func sum(p []uint64) uint64 {
+	var a, b, c, d, e, f, g, h uint64
+	i := 0
+	for ; i+8 <= len(p); i += 8 {
+		a += p[i]
+		b += p[i+1]
+		c += p[i+2]
+		d += p[i+3]
+		e += p[i+4]
+		f += p[i+5]
+		g += p[i+6]
+		h += p[i+7]
+	}
+	for ; i < len(p); i++ {
+		a += p[i]
+	}
+	return a + b + c + d + e + f + g + h
+}
+
+// sinkValue keeps the sums alive, so the reads are not optimised away.
+var sinkValue struct {
+	sync.Mutex
+	v uint64
+}
+
+func sink(v uint64) {
+	sinkValue.Lock()
+	sinkValue.v += v
+	sinkValue.Unlock()
+}
+
+// memAvailable is MemAvailable in bytes, 0 when unknown.
+func memAvailable(path string) int64 {
+	f, err := os.Open(path) //nolint:gosec // a fixed system path
+	if err != nil {
+		return 0
+	}
+	defer f.Close()
+	sc := bufio.NewScanner(f)
+	for sc.Scan() {
+		fields := strings.Fields(sc.Text())
+		if len(fields) >= 2 && fields[0] == "MemAvailable:" {
+			kb, _ := strconv.ParseInt(fields[1], 10, 64)
+			return kb << 10
+		}
+	}
+	return 0
+}
+
+// lastCacheBytes is the largest cache any CPU reports, 0 when unknown.
+func lastCacheBytes(root string) int64 {
+	paths, _ := filepath.Glob(filepath.Join(root, "cpu*", "cache", "index*", "size"))
+	var most int64
+	for _, p := range paths {
+		b, err := os.ReadFile(p) //nolint:gosec // under /sys
+		if err != nil {
+			continue
+		}
+		v := strings.TrimSpace(string(b))
+		mult := int64(1)
+		switch {
+		case strings.HasSuffix(v, "K"):
+			mult, v = 1<<10, strings.TrimSuffix(v, "K")
+		case strings.HasSuffix(v, "M"):
+			mult, v = 1<<20, strings.TrimSuffix(v, "M")
+		}
+		if n, err := strconv.ParseInt(v, 10, 64); err == nil {
+			most = max(most, n*mult)
+		}
+	}
+	return most
+}

internal/agent/discovery/membw_test.go

@@ -0,0 +1,32 @@
+package discovery
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+func TestMemoryBandwidthIsMeasured(t *testing.T) {
+	if testing.Short() {
+		t.Skip("reads half a gigabyte")
+	}
+	bw := MeasureMemoryBandwidth()
+	// Any machine running this reads more than a gigabyte a second and less
+	// than a few terabytes.
+	if bw < 1 || bw > 5000 {
+		t.Errorf("measured %.1f GB/s", bw)
+	}
+	t.Logf("%.1f GB/s", bw)
+}
+
+func TestTheLastCacheIsReadFromSys(t *testing.T) {
+	root := t.TempDir()
+	for path, size := range map[string]string{"cpu0/cache/index0/size": "48K", "cpu0/cache/index3/size": "96M"} {
+		p := filepath.Join(root, path)
+		_ = os.MkdirAll(filepath.Dir(p), 0o755)
+		_ = os.WriteFile(p, []byte(size+"\n"), 0o644)
+	}
+	if got := lastCacheBytes(root); got != 96<<20 {
+		t.Errorf("%d", got)
+	}
+}

internal/agent/discovery/pcie.go

@@ -0,0 +1,181 @@
+package discovery
+
+import (
+	"context"
+	"encoding/csv"
+	"fmt"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strconv"
+	"strings"
+)
+
+// A card in the wrong slot: an x16 card in a slot wired x8 or x4, behind a
+// riser, or in a slot of an older PCIe generation than the card's, links
+// slower than it was built to, and nothing says so. A model on the card
+// still decodes at the card's own memory speed, but it loads slower, a
+// tensor-parallel replica over PCIe is slower, and so is anything that
+// spills to the host. The driver says how the link stands and what the
+// card could do; the difference is reported where the node's warnings are.
+
+// PCIeLink is how a card is linked to its host.
+type PCIeLink struct {
+	// Gen is the fastest generation the card and the slot together allow —
+	// not the current one, which drops while the card idles, to save power.
+	// GPUGen is the card's own fastest, 0 when the driver does not say.
+	Gen    int `json:"gen,omitempty"`
+	GPUGen int `json:"gpu_gen,omitempty"`
+	// Width is the lanes in use now, WidthMax the card's most.
+	Width    int `json:"width,omitempty"`
+	WidthMax int `json:"width_max,omitempty"`
+}
+
+// laneGBs is one lane's usable rate in GB/s, by generation.
+var laneGBs = map[int]float64{1: 0.25, 2: 0.5, 3: 0.985, 4: 1.969, 5: 3.938, 6: 7.877}
+
+// GBs is what a link of gen and width carries, in GB/s each way.
+func pcieGBs(gen, width int) float64 { return laneGBs[gen] * float64(width) }
+
+// Problem says how a card's link falls short of what the card could do,
+// empty when it does not or the driver did not say.
+func (p PCIeLink) Problem() string {
+	gen := p.Gen
+	if p.GPUGen > 0 && gen == 0 {
+		gen = p.GPUGen
+	}
+	slowGen := p.GPUGen > 0 && p.Gen > 0 && p.Gen < p.GPUGen
+	narrow := p.WidthMax > 0 && p.Width > 0 && p.Width < p.WidthMax
+	if !slowGen && !narrow {
+		return ""
+	}
+	could := pcieGBs(max(p.GPUGen, gen), p.WidthMax)
+	has := pcieGBs(gen, p.Width)
+	var why []string
+	if narrow {
+		why = append(why, fmt.Sprintf("x%d of its x%d lanes", p.Width, p.WidthMax))
+	}
+	if slowGen {
+		why = append(why, fmt.Sprintf("PCIe Gen%d of its Gen%d", p.Gen, p.GPUGen))
+	}
+	return fmt.Sprintf("linked at %s: %.1f GB/s where it could have %.1f. "+
+		"Check the slot (wired fewer lanes, or an older generation), a riser, or the BIOS's PCIe setting",
+		strings.Join(why, " and "), has, could)
+}
+
+// pcieFields are asked apart from the rest, and the card's own fastest
+// generation apart again: a driver that does not know a field refuses the
+// whole query, and these must not cost a node its cards.
+var pcieFields = [][]string{
+	{"index", "pcie.link.gen.max", "pcie.link.width.current", "pcie.link.width.max", "pcie.link.gen.gpumax"},
+	{"index", "pcie.link.gen.max", "pcie.link.width.current", "pcie.link.width.max"},
+}
+
+// readPCIe fills each card's link from nvidia-smi, leaving it empty when
+// the driver says nothing of it.
+func readPCIe(ctx context.Context, binary string, cards []GPU) {
+	for _, fields := range pcieFields {
+		out, err := exec.CommandContext(ctx, binary, "--query-gpu="+strings.Join(fields, ","),
+			"--format=csv,noheader,nounits").Output()
+		if err != nil {
+			continue
+		}
+		links := parsePCIe(string(out))
+		for i := range cards {
+			if l, ok := links[cards[i].Index]; ok && cards[i].MIG == nil {
+				cards[i].PCIe = l
+			}
+		}
+		break
+	}
+	for i := range cards {
+		if cards[i].MIG == nil {
+			cards[i].PCIe = withSysfs(cards[i].PCIe, "/sys/bus/pci/devices", cards[i].PCIBusID)
+		}
+	}
+}
+
+// withSysfs takes the card's own lane count, and its own fastest
+// generation, from the kernel: the driver's width.max is what card and slot
+// allow together, so a x16 card in a slot wired x8 reads x8 of x8 there,
+// while the device's max_link_width is the card's, 16.
+func withSysfs(l PCIeLink, root, busID string) PCIeLink {
+	dev := sysfsBusID(busID)
+	if dev == "" {
+		return l
+	}
+	read := func(name string) string {
+		b, err := os.ReadFile(filepath.Join(root, dev, name)) //nolint:gosec // under /sys
+		if err != nil {
+			return ""
+		}
+		return strings.TrimSpace(string(b))
+	}
+	if n, err := strconv.Atoi(read("max_link_width")); err == nil && n > 0 {
+		l.WidthMax = n
+	}
+	if n, err := strconv.Atoi(read("current_link_width")); err == nil && n > 0 {
+		l.Width = n
+	}
+	if g := genOfSpeed(read("max_link_speed")); g > 0 && l.GPUGen == 0 {
+		l.GPUGen = g
+	}
+	return l
+}
+
+// sysfsBusID is nvidia-smi's 00000000:3B:00.0 as the kernel names it,
+// 0000:3b:00.0.
+func sysfsBusID(id string) string {
+	id = strings.ToLower(strings.TrimSpace(id))
+	dom, rest, ok := strings.Cut(id, ":")
+	if !ok || len(dom) < 4 {
+		return ""
+	}
+	return dom[len(dom)-4:] + ":" + rest
+}
+
+// genOfSpeed is the PCIe generation of a link speed as sysfs writes it.
+func genOfSpeed(s string) int {
+	switch {
+	case strings.HasPrefix(s, "2.5 GT/s"):
+		return 1
+	case strings.HasPrefix(s, "5.0 GT/s"), strings.HasPrefix(s, "5 GT/s"):
+		return 2
+	case strings.HasPrefix(s, "8.0 GT/s"), strings.HasPrefix(s, "8 GT/s"):
+		return 3
+	case strings.HasPrefix(s, "16.0 GT/s"), strings.HasPrefix(s, "16 GT/s"):
+		return 4
+	case strings.HasPrefix(s, "32.0 GT/s"), strings.HasPrefix(s, "32 GT/s"):
+		return 5
+	case strings.HasPrefix(s, "64.0 GT/s"), strings.HasPrefix(s, "64 GT/s"):
+		return 6
+	}
+	return 0
+}
+
+func parsePCIe(out string) map[int]PCIeLink {
+	r := csv.NewReader(strings.NewReader(out))
+	r.TrimLeadingSpace = true
+	r.FieldsPerRecord = -1
+	rows, _ := r.ReadAll()
+	num := func(s string) int {
+		n, _ := strconv.Atoi(strings.TrimSpace(s))
+		return n
+	}
+	links := map[int]PCIeLink{}
+	for _, row := range rows {
+		if len(row) < 4 {
+			continue
+		}
+		idx, err := strconv.Atoi(strings.TrimSpace(row[0]))
+		if err != nil {
+			continue
+		}
+		l := PCIeLink{Gen: num(row[1]), Width: num(row[2]), WidthMax: num(row[3])}
+		if len(row) > 4 {
+			l.GPUGen = num(row[4])
+		}
+		links[idx] = l
+	}
+	return links
+}

internal/agent/discovery/pcie_test.go

@@ -0,0 +1,54 @@
+package discovery
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestACardInTheWrongSlotIsSaidSo(t *testing.T) {
+	for _, c := range []struct {
+		link PCIeLink
+		want string // empty: nothing to say
+	}{
+		// The PoC's Quadro RTX 6000, idle: current Gen1, but its fastest is 3,
+		// as the card's is, on all 16 lanes. Nothing wrong.
+		{PCIeLink{Gen: 3, GPUGen: 3, Width: 16, WidthMax: 16}, ""},
+		// An x16 card in a slot wired x8.
+		{PCIeLink{Gen: 4, GPUGen: 4, Width: 8, WidthMax: 16}, "x8 of its x16 lanes: 15.8 GB/s where it could have 31.5"},
+		// A Gen5 card in a Gen4 slot.
+		{PCIeLink{Gen: 4, GPUGen: 5, Width: 16, WidthMax: 16}, "PCIe Gen4 of its Gen5: 31.5 GB/s where it could have 63.0"},
+		// Both, behind a riser.
+		{PCIeLink{Gen: 3, GPUGen: 4, Width: 4, WidthMax: 16}, "x4 of its x16 lanes and PCIe Gen3 of its Gen4"},
+		// A driver that says nothing.
+		{PCIeLink{}, ""},
+	} {
+		got := c.link.Problem()
+		if (c.want == "") != (got == "") || !strings.Contains(got, c.want) {
+			t.Errorf("%+v: %q, want %q", c.link, got, c.want)
+		}
+	}
+}
+
+func TestTheLinkIsReadFromNvidiaSMI(t *testing.T) {
+	links := parsePCIe("0, 3, 16, 16, 3\n1, 4, 8, 16, [N/A]\n")
+	if links[0] != (PCIeLink{Gen: 3, GPUGen: 3, Width: 16, WidthMax: 16}) || links[1] != (PCIeLink{Gen: 4, Width: 8, WidthMax: 16}) {
+		t.Errorf("%+v", links)
+	}
+}
+
+func TestTheCardsOwnLanesComeFromTheKernel(t *testing.T) {
+	// The driver says x8 of x8 for a x16 card in a slot wired x8; the
+	// device's own max_link_width says 16.
+	root := t.TempDir()
+	dev := filepath.Join(root, "0000:3b:00.0")
+	_ = os.MkdirAll(dev, 0o755)
+	for f, v := range map[string]string{"max_link_width": "16", "current_link_width": "8", "max_link_speed": "16.0 GT/s PCIe"} {
+		_ = os.WriteFile(filepath.Join(dev, f), []byte(v+"\n"), 0o644)
+	}
+	l := withSysfs(PCIeLink{Gen: 4, Width: 8, WidthMax: 8}, root, "00000000:3B:00.0")
+	if l.WidthMax != 16 || l.Width != 8 || l.GPUGen != 4 || !strings.Contains(l.Problem(), "x8 of its x16") {
+		t.Errorf("%+v: %q", l, l.Problem())
+	}
+}

internal/agent/engines/engine.go

@@ -47,6 +47,9 @@ type Profile struct {
 	// KVEvents has the engine publish its KV cache events to the agent,
 	// for the router's precise index. vLLM only.
 	KVEvents bool `json:"kv_events,omitempty"`
+	// AllowPickle lets the node fetch weights in a pickle format, which run
+	// code when loaded; the catalog's allow_pickle.
+	AllowPickle bool `json:"allow_pickle,omitempty"`
 }
 
 // ShardSpec is one shard of one replica on this node.
@@ -215,6 +218,7 @@ func NewRegistry(cfg Config) *Registry {
 	r.Add(&TRTLLM{cfg: cfg.TRTLLM})
 	r.Add(&Sim{cfg: cfg.Sim})
 	r.Add(&ComfyUI{cfg: cfg.ComfyUI})
+	r.Add(&Strata{cfg: cfg.Strata})
 	return r
 }
 
@@ -265,6 +269,9 @@ type Config struct {
 	Sim    SimConfig    `koanf:"sim"`
 	// ComfyUI makes pictures; offered by a node whose ComfyUI is named here.
 	ComfyUI ComfyUIConfig `koanf:"comfyui"`
+	// Strata serves a few mixture-of-experts models from RAM and one card;
+	// offered by a node whose Strata is named here, or that installed it.
+	Strata StrataConfig `koanf:"strata"`
 }
 
 // --- shared helpers ---

internal/agent/engines/engines_test.go

@@ -477,14 +477,14 @@ func TestValidateRequiresModelPath(t *testing.T) {
 
 func TestRegistry(t *testing.T) {
 	r := NewRegistry(Config{})
-	if got := r.Names(); !slices.Equal(got, []string{"comfyui", "llamacpp", "sglang", "sim", "trtllm", "vllm"}) {
+	if got := r.Names(); !slices.Equal(got, []string{"comfyui", "llamacpp", "sglang", "sim", "strata", "trtllm", "vllm"}) {
 		t.Errorf("registry contents: %v", got)
 	}
 	if _, err := r.Get("vllm"); err != nil {
 		t.Errorf("vllm missing: %v", err)
 	}
 	err := func() error { _, err := r.Get("nope"); return err }()
-	if err == nil || !strings.Contains(err.Error(), "llamacpp, sglang, sim, trtllm, vllm") {
+	if err == nil || !strings.Contains(err.Error(), "llamacpp, sglang, sim, strata, trtllm, vllm") {
 		t.Errorf("an unknown engine should list what is available: %v", err)
 	}
 }

internal/agent/engines/pickle_test.go

@@ -0,0 +1,58 @@
+package engines
+
+import (
+	"slices"
+	"testing"
+)
+
+func TestAModelDownloadedByTheEngineIsReadAsSafetensorsOnly(t *testing.T) {
+	for path, want := range map[string]bool{
+		"Qwen/Qwen2.5-7B-Instruct":    true,
+		"/var/lib/llmgw/weights/abc/": false, "./local": false,
+		"http://m/q/v/": false, "q.gguf": false,
+	} {
+		got := slices.Contains(safetensorsOnly(Profile{ModelPath: path}), "safetensors")
+		if got != want {
+			t.Errorf("%q: safetensors only = %v, want %v", path, got, want)
+		}
+	}
+	if len(safetensorsOnly(Profile{ModelPath: "Qwen/Qwen2.5-7B-Instruct", AllowPickle: true})) != 0 {
+		t.Error("allow_pickle still forced safetensors")
+	}
+}
+
+func TestAForbiddenFlagIsKnownHoweverItIsSpelled(t *testing.T) {
+	for engine, args := range map[string][]string{
+		"vllm":   {"--trust_remote_code"},
+		"sglang": {"--load-format", "pt"},
+	} {
+		if UnsafeArgs(engine, args) == nil {
+			t.Errorf("%s %v was let through", engine, args)
+		}
+	}
+	for _, a := range [][]string{{"--load_format", "pt"}, {"--trust-remote"}, {"--load-form=pt"}} {
+		if UnsafeArgs("vllm", a) == nil {
+			t.Errorf("vllm %v was let through", a)
+		}
+	}
+	// What is not one of them, nor the start of one, still passes.
+	for _, a := range [][]string{{"--max-model-len", "8192"}, {"--tokenizer-mode", "mistral"}, {"--gpu-memory-utilization=0.9"}} {
+		if err := UnsafeArgs("vllm", a); err != nil {
+			t.Errorf("vllm %v was refused: %v", a, err)
+		}
+	}
+	// Short flags are matched as written; llama.cpp takes no abbreviations,
+	// so its --chat-template is not taken for --chat-template-file.
+	for _, a := range [][]string{{"-m", "x"}, {"-hf", "o/r"}, {"-mu", "http://x"}} {
+		if UnsafeArgs("llamacpp", a) == nil {
+			t.Errorf("llamacpp %v was let through", a)
+		}
+	}
+	if err := UnsafeArgs("llamacpp", []string{"--chat-template", "chatml"}); err != nil {
+		t.Errorf("llama.cpp's own --chat-template was refused: %v", err)
+	}
+	// ComfyUI's argparse reads --li as --listen.
+	if UnsafeArgs("comfyui", []string{"--li", "0.0.0.0"}) == nil {
+		t.Error("comfyui --li was let through")
+	}
+}

internal/agent/engines/sglang.go

@@ -102,6 +102,7 @@ func (g *SGLang) serverCommand(s ShardSpec) (Command, error) {
 	if p.ServedModelName != "" {
 		args = append(args, "--served-model-name", p.ServedModelName)
 	}
+	args = append(args, safetensorsOnly(p)...)
 	if s.NNodes > 1 {
 		args = append(args,
 			"--nnodes", strconv.Itoa(s.NNodes),

internal/agent/engines/strata.go

@@ -0,0 +1,144 @@
+package engines
+
+import (
+	"fmt"
+	"net"
+	"os"
+	"path/filepath"
+	"strconv"
+	"sync/atomic"
+)
+
+// StrataConfig locates a Strata installation.
+type StrataConfig struct {
+	// Dir is Strata's source directory, where setup.py and serve/ are.
+	Dir string `koanf:"dir"`
+	// Python is the interpreter of its environment; Dir's .venv when unset.
+	Python string `koanf:"python"`
+	// Engine is the directory of its prebuilt engine, where the strata
+	// binary and the CUDA libraries it ships are; Dir's engine/ when unset.
+	Engine string `koanf:"engine"`
+}
+
+// Strata runs Strata (github.com/Niko1221/Strata): a mixture-of-experts
+// engine that keeps a model's experts in RAM and the ones used most in VRAM,
+// so a 125B model serves from one 24 GB card. It runs a few model families
+// only, each prepared from its GGUF files by Strata's own setup, so the agent
+// starts llmgw strata-serve, which prepares the model and then becomes
+// Strata's server.
+type Strata struct {
+	cfg       StrataConfig
+	installed atomic.Pointer[string]
+}
+
+// Name identifies the engine in the catalog.
+func (*Strata) Name() string { return "strata" }
+
+// HealthPath answers 200 only once the model is loaded.
+func (*Strata) HealthPath() string { return "/health" }
+
+// MetricsPath is Strata's Prometheus text, in vLLM's names; its /metrics is
+// JSON for its own monitor unless asked for this format.
+func (*Strata) MetricsPath() string { return "/metrics?format=prometheus" }
+
+// ServesHTTP is true: a Strata replica is one process on one node.
+func (*Strata) ServesHTTP(ShardSpec) bool { return true }
+
+// UseInstalled points the engine at the environment the agent installed:
+// its interpreter, with Strata's code beside its bin in strata/ and the
+// engine in strata-engine/.
+func (s *Strata) UseInstalled(python string) { s.installed.Store(&python) }
+
+// Configured reports whether the configuration names a Strata.
+func (s *Strata) Configured() bool { return s.cfg.Dir != "" }
+
+// StrataSourceDir and StrataEngineDir are where an installed environment
+// keeps Strata's code and its prebuilt engine, beside bin/.
+const (
+	StrataSourceDir = "strata"
+	StrataEngineDir = "strata-engine"
+)
+
+func (s *Strata) env() string {
+	if p := s.installed.Load(); p != nil {
+		return filepath.Dir(filepath.Dir(*p))
+	}
+	return ""
+}
+
+func (s *Strata) dir() string {
+	if e := s.env(); e != "" {
+		return filepath.Join(e, StrataSourceDir)
+	}
+	return s.cfg.Dir
+}
+
+func (s *Strata) engineDir() string {
+	if e := s.env(); e != "" {
+		return filepath.Join(e, StrataEngineDir)
+	}
+	if s.cfg.Engine != "" {
+		return s.cfg.Engine
+	}
+	return filepath.Join(s.cfg.Dir, "engine")
+}
+
+func (s *Strata) python() string {
+	if p := s.installed.Load(); p != nil {
+		return *p
+	}
+	if s.cfg.Python != "" {
+		return s.cfg.Python
+	}
+	return filepath.Join(s.cfg.Dir, ".venv", "bin", "python")
+}
+
+// Validate checks the profile: one card, on one node. Strata splits a model
+// over a node's cards by layer, but that is not driven from here yet.
+func (*Strata) Validate(p Profile, gpus []GPUInfo) error {
+	if p.Nodes > 1 || p.TP > 1 || p.PP > 1 || p.GPUsPerNode > 1 {
+		return fmt.Errorf("profile %q: strata runs on one GPU of one node", p.Name)
+	}
+	if err := requireComputeCapability(p, gpus, 7.0, "strata"); err != nil {
+		return err
+	}
+	return validateCommon(p, gpus)
+}
+
+// Commands is llmgw strata-serve, which prepares the model with Strata's
+// setup and then execs Strata's server.
+func (s *Strata) Commands(sh ShardSpec) ([]Command, error) {
+	dir := s.dir()
+	if dir == "" {
+		return nil, fmt.Errorf("strata: no Strata directory is configured (engines.strata.dir)")
+	}
+	bin, err := os.Executable()
+	if err != nil {
+		return nil, err
+	}
+	p := sh.Profile
+	args := []string{"strata-serve",
+		"--listen", net.JoinHostPort(sh.Host(), strconv.Itoa(sh.HTTPPort)),
+		"--model", p.ServedModelName,
+		"--weights", p.ModelPath,
+		// Relative to where engines run — their user's home — so they are
+		// theirs to write. The data is shared by the node's replicas: a
+		// prepared model and its draft layer are gigabytes, the same for
+		// every replica of it.
+		"--state", filepath.Join("strata", sh.ReplicaID),
+		"--data", filepath.Join("strata", "data"),
+		"--python", s.python(),
+		"--strata", dir,
+		"--engine", s.engineDir(),
+		"--",
+	}
+	args = append(args, p.Args...)
+	env := baseEnv(sh)
+	if sh.APIKey != "" {
+		// Strata's server reads its key from here, and strata-serve passes
+		// it on in the environment: not on a command line every user on the
+		// node can read.
+		env["STRATA_API_KEY"] = sh.APIKey
+	}
+	return []Command{{Name: "strata", Path: bin, Args: args, Env: materializeEnv(env)}}, nil
+}

internal/agent/engines/strata_test.go

@@ -0,0 +1,69 @@
+package engines
+
+import (
+	"path/filepath"
+	"slices"
+	"strings"
+	"testing"
+)
+
+func TestStrataStartsStrataServeWithTheKeyInTheEnvironment(t *testing.T) {
+	s := &Strata{}
+	s.UseInstalled("/opt/llmgw/engines/strata/0.2.7-cuda13/bin/python")
+	spec := ShardSpec{
+		ReplicaID: "r1", HTTPPort: 18090, BindAddr: "10.0.0.147", GPUIndices: []int{0},
+		APIKey: "secret-key",
+		Profile: Profile{Name: "p", Engine: "strata", TP: 1, ModelPath: "/var/lib/llmgw/models/q38",
+			ServedModelName: "qwen3.8-flash-next", Args: []string{"--family", "qwen", "--size", "IQ2_XS"}},
+	}
+	cmds, err := s.Commands(spec)
+	if err != nil {
+		t.Fatal(err)
+	}
+	c := cmds[0]
+	args := strings.Join(c.Args, " ")
+	env := "/opt/llmgw/engines/strata/0.2.7-cuda13"
+	for _, want := range []string{
+		"strata-serve", "--listen 10.0.0.147:18090", "--model qwen3.8-flash-next",
+		"--weights /var/lib/llmgw/models/q38", "--state " + filepath.Join("strata", "r1"),
+		"--python " + env + "/bin/python", "--strata " + env + "/strata", "--engine " + env + "/strata-engine",
+		"-- --family qwen --size IQ2_XS",
+	} {
+		if !strings.Contains(args, want) {
+			t.Errorf("args lack %q: %s", want, args)
+		}
+	}
+	if strings.Contains(args, "secret-key") {
+		t.Error("the key is on the command line, which every user on the node can read")
+	}
+	if !slices.Contains(c.Env, "STRATA_API_KEY=secret-key") {
+		t.Error("the key is not in the environment, where Strata's server reads it")
+	}
+	if !slices.Contains(c.Env, "CUDA_VISIBLE_DEVICES=0") {
+		t.Error("the shard is not pinned to its GPU")
+	}
+}
+
+func TestStrataRunsOnOneCard(t *testing.T) {
+	s := &Strata{}
+	p := Profile{Name: "p", Engine: "strata", TP: 1, GPUsPerNode: 2, ModelPath: "/m"}
+	if err := s.Validate(p, nil); err == nil {
+		t.Error("a two-card profile was accepted")
+	}
+	p.GPUsPerNode = 1
+	if err := s.Validate(p, []GPUInfo{{Name: "Quadro RTX 6000", ComputeCapability: "7.5"}}); err != nil {
+		t.Errorf("one Turing card was refused: %v", err)
+	}
+}
+
+func TestStrataSetupFlagsThatMoveFilesAreRefused(t *testing.T) {
+	for _, bad := range []string{"--gguf-dir", "--gguf", "--data-dir=/etc", "--model", "--mod", "--feed", "--host", "--gpus"} {
+		if err := UnsafeArgs("strata", []string{bad, "x"}); err == nil {
+			t.Errorf("%s was accepted", bad)
+		}
+	}
+	if err := UnsafeArgs("strata", []string{"--family", "qwen", "--size", "IQ2_XS", "--context", "32768",
+		"--parallel", "2", "--vram-reserve-mib", "2048", "--vision", "no"}); err != nil {
+		t.Errorf("an ordinary profile was refused: %v", err)
+	}
+}

internal/agent/engines/unsafeenv.go

@@ -2,6 +2,7 @@ package engines
 
 import (
 	"fmt"
+	"os"
 	"strings"
 )
 
@@ -101,6 +102,13 @@ var flagsNever = map[string]map[string]string{
 		"--hf-token": "is a credential", "--alias": "is set by the agent",
 		"--model-draft": "loads a file", "-md": "loads a file",
 		"--prompt-cache": "writes a file", "--logdir": "writes files",
+		// The short spellings of the same.
+		"-mm": "loads a file; the projector is fetched with the model", "--mmproj-url": "downloads a file",
+		"-mmu": "downloads a file", "-hfr": "downloads a model", "-hff": "downloads a model",
+		"--hf-repo-draft": "downloads a model", "-hfd": "downloads a model", "-hfrd": "downloads a model",
+		"--hf-repo-v": "downloads a model", "-hfv": "downloads a model", "--hf-file-v": "downloads a model",
+		"-hffv": "downloads a model", "-hft": "is a credential", "-jf": "reads a file", "-a": "is set by the agent",
+		"--model-vocoder": "loads a file", "-mv": "loads a file",
 	},
 	"vllm": {
 		"--download-dir": "writes files", "--tokenizer": "reads another repository or path",
@@ -118,6 +126,7 @@ var flagsNever = map[string]map[string]string{
 		"--file-storage-path": "writes files", "--log-requests": "logs prompts",
 		"--model-path": "is the model's source, set by the agent", "--served-model-name": "is set by the agent",
 		"--lora-paths": "loads files", "--download-dir": "writes files",
+		"--load-format": "changes how weights are loaded, some ways running code",
 	},
 	// ComfyUI's main.py, behind the gate that sets where it listens and
 	// where its files are.
@@ -132,6 +141,22 @@ var flagsNever = map[string]map[string]string{
 		"--tls-keyfile":        "reads a file", "--tls-certfile": "reads a file",
 		"--comfy-api-base": "sends work to another service",
 	},
+	// Strata's setup.py, which a profile's arguments go to: llmgw
+	// strata-serve sets where the model is, where its files go and where
+	// the server listens, and the engine is the one the agent installed.
+	// Its own --model, the size, is spelt --size in a profile, since
+	// --model is the source everywhere else.
+	"strata": {
+		"--gguf-dir": "is the model's source, set by the agent", "--models-dir": "is set by the agent",
+		"--data-dir": "is set by the agent", "--move-data": "moves files", "--set-default-data-dir": "writes settings",
+		"--feed": "downloads an engine", "--prebuilt": "loads an engine", "--build": "compiles an engine",
+		"--update": "downloads an engine", "--rollback-engine": "changes the engine",
+		"--dsv41-package": "loads a file", "--inspect": "reads another repository or path",
+		"--source": "downloads from elsewhere", "--setup": "is not for a server",
+		"--browser": "opens a browser", "--gpu": "is set by the agent", "--gpus": "is set by the agent",
+		"--gpu-roles": "is set by the agent", "--backend": "is set by the agent",
+		"--experimental-speed-projection": "changes what the model refuses",
+	},
 	// trtllm-serve spells its flags with underscores, so the shared list's
 	// hyphenated names do not reach them.
 	"trtllm": {
@@ -163,14 +188,43 @@ var flagsNever = map[string]map[string]string{
 // unsafeFlag is why an engine may not be given a flag, or "".
 func unsafeFlag(engine, flag, _ string) string {
 	name, _, _ := strings.Cut(flag, "=")
+	if !strings.HasPrefix(name, "-") {
+		return ""
+	}
+	// Written as the engine would read it: vLLM takes --load_format for
+	// --load-format, TensorRT-LLM the other way round. Short flags (-m,
+	// -hf) are matched as written.
+	norm := func(f string) string {
+		if strings.HasPrefix(f, "--") {
+			return "--" + strings.ReplaceAll(f[2:], "_", "-")
+		}
+		return f
+	}
+	given := norm(name)
+	// argparse also takes any unambiguous beginning of a long flag for the
+	// flag — --trust-remote is --trust-remote-code, --li is ComfyUI's
+	// --listen — so for the engines that parse with it, a beginning of a
+	// refused flag is refused. llama.cpp and TensorRT-LLM take no such
+	// thing, and --chat-template there is a flag of its own.
+	abbreviates := argparseEngines[engine] && strings.HasPrefix(given, "--") && len(given) >= 4
 	for _, set := range []string{"", engine} {
-		if why, ok := flagsNever[set][name]; ok && why != "" {
-			return why
+		for never, why := range flagsNever[set] {
+			if why == "" {
+				continue
+			}
+			n := norm(never)
+			if n == given || (abbreviates && strings.HasPrefix(n, "--") && strings.HasPrefix(n, given)) {
+				return why
+			}
 		}
 	}
 	return ""
 }
 
+// argparseEngines read their flags with Python's argparse, which takes a
+// long flag's beginning for the flag.
+var argparseEngines = map[string]bool{"vllm": true, "sglang": true, "comfyui": true, "strata": true}
+
 // UnsafeArgs refuses the first of a profile's arguments an engine may not be
 // given, saying why.
 func UnsafeArgs(engine string, args []string) error {
@@ -185,3 +239,29 @@ func UnsafeArgs(engine string, args []string) error {
 	}
 	return nil
 }
+
+// safetensorsOnly has vLLM or SGLang read only safetensors from a model it
+// downloads itself, by its Hugging Face id: left to choose, either takes
+// the repository's pickle files (pytorch_model.bin) when it has no
+// safetensors, and loading those runs whatever they hold. A path was
+// fetched by the agent, which checked it, or put there by the operator.
+func safetensorsOnly(p Profile) []string {
+	if p.AllowPickle || !hubID(p.ModelPath) {
+		return nil
+	}
+	return []string{"--load-format", "safetensors"}
+}
+
+// hubID reports whether a model path is a Hugging Face id, org/name, rather
+// than a file or directory.
+func hubID(s string) bool {
+	if s == "" || strings.HasPrefix(s, "/") || strings.HasPrefix(s, ".") || strings.Contains(s, "://") {
+		return false
+	}
+	// org/name:Q4_K_M is a GGUF in a repository, which vLLM reads as one.
+	if strings.Count(s, "/") != 1 || strings.Contains(s, ":") {
+		return false
+	}
+	_, err := os.Stat(s)
+	return err != nil
+}

internal/agent/engines/vllm.go

@@ -119,6 +119,7 @@ func (v *VLLM) serverCommand(s ShardSpec) (Command, error) {
 	if p.ServedModelName != "" {
 		args = append(args, "--served-model-name", p.ServedModelName)
 	}
+	args = append(args, safetensorsOnly(p)...)
 
 	env := baseEnv(s)
 

internal/agent/gpucheck.go

@@ -20,6 +20,7 @@ var (
 	gpuProcesses = discovery.GPUProcesses
 	gpuMemory    = discovery.GPUMemoryByProcess
 	processGroup = pgidOf
+	parentOf     = ppidOf
 	// gpuCheckWait is how long a shard has, once it answers, to show up on
 	// its GPUs: the weights are on them by then, but nvidia-smi can lag.
 	gpuCheckWait = 15 * time.Second
@@ -116,19 +117,44 @@ func (a *Agent) foreignGPUMemory(ctx context.Context, shards []int) (foreign, us
 	return foreign, used
 }
 
-// holds reports whether any of pids is the shard's process or in its group:
-// an engine that computes in a child process is still the shard.
+// holds reports whether any of pids is the shard's process, in its group,
+// or a process it started: an engine that computes in a child process is
+// still the shard, and Strata's server starts its engine in a group of the
+// engine's own.
 func holds(pids []int, pid, group int) bool {
 	for _, p := range pids {
-		if p == pid || (group > 0 && processGroup(p) == group) {
+		if p == pid || (group > 0 && processGroup(p) == group) || descends(p, pid) {
 			return true
 		}
 	}
 	return false
 }
 
+// descends reports whether process p was started, at any remove, by pid.
+func descends(p, pid int) bool {
+	// Bounded: a chain this deep is no engine's, and a /proc read racing an
+	// exit must not loop.
+	for range 16 {
+		p = parentOf(p)
+		switch {
+		case p == pid:
+			return true
+		case p <= 1:
+			return false
+		}
+	}
+	return false
+}
+
 // pgidOf reads a process's group from /proc; zero when it cannot.
-func pgidOf(pid int) int {
+func pgidOf(pid int) int { return statField(pid, 2) }
+
+// ppidOf reads a process's parent from /proc; zero when it cannot.
+func ppidOf(pid int) int { return statField(pid, 1) }
+
+// statField is one of the numbers after the command name in /proc's stat
+// line: 1 is the parent, 2 the group.
+func statField(pid, field int) int {
 	b, err := os.ReadFile("/proc/" + strconv.Itoa(pid) + "/stat")
 	if err != nil {
 		return 0
@@ -141,10 +167,10 @@ func pgidOf(pid int) int {
 		return 0
 	}
 	f := strings.Fields(s[i+1:])
-	if len(f) < 3 {
+	if len(f) <= field {
 		return 0
 	}
-	n, _ := strconv.Atoi(f[2])
+	n, _ := strconv.Atoi(f[field])
 	return n
 }
 

internal/agent/gpucheck_test.go

@@ -123,6 +123,30 @@ func TestForeignGPUMemoryLeavesOutTheAgentsOwnShards(t *testing.T) {
 	}
 }
 
+func TestAnEngineStartedInAGroupOfItsOwnIsStillTheShards(t *testing.T) {
+	// cuda01: Strata's server (259709) starts its engine (261072) in a
+	// group of the engine's own, and the engine holds the card. It read as
+	// a shard on the CPU, and as 22 GB of somebody else's memory.
+	oldMem, oldGroup, oldParent := gpuMemory, processGroup, parentOf
+	t.Cleanup(func() { gpuMemory, processGroup, parentOf = oldMem, oldGroup, oldParent })
+	processGroup = func(pid int) int { return pid }
+	parentOf = func(pid int) int {
+		return map[int]int{261072: 259709, 259709: 258734, 258734: 1, 999: 1}[pid]
+	}
+	gpuMemory = func(context.Context, string) ([]discovery.GPUMemory, error) {
+		return []discovery.GPUMemory{{Index: 0, UsedMB: 23000,
+			Procs: []discovery.GPUProcess{{PID: 261072, UsedMB: 22226}, {PID: 999, UsedMB: 700}}}}, nil
+	}
+	a := &Agent{log: slog.Default()}
+	if c, ok := a.gpuCompute(context.Background(), []int{0}, 259709); !ok || c.Backend != "CUDA" ||
+		len(c.Devices) != 1 || c.Devices[0] != "GPU 0: 21.7 GB" {
+		t.Errorf("the engine's child on GPU 0: %+v %v", c, ok)
+	}
+	if foreign, _ := a.foreignGPUMemory(context.Background(), []int{259709}); !slices.Equal(foreign, []int32{774}) {
+		t.Errorf("foreign %v; want only what the other process and nobody holds, [774]", foreign)
+	}
+}
+
 func TestLlamaCppsOwnWordOnTheCPUIsEnoughToRefuse(t *testing.T) {
 	eng := fallbackEngine{}
 	// Without nvidia-smi to ask, the engine's own output says it.

internal/agent/main_test.go

@@ -0,0 +1,15 @@
+package agent
+
+import (
+	"os"
+	"testing"
+	"time"
+)
+
+func TestMain(m *testing.M) {
+	// Half a gigabyte read per test binary, and contending with the tests
+	// for the CPUs: these never look at the figure.
+	measureBandwidth = func() float64 { return 0 }
+	bandwidthDelay = func() time.Duration { return 0 }
+	os.Exit(m.Run())
+}

internal/agent/provision/install.go

@@ -174,7 +174,7 @@ func (in *Installer) one(ctx context.Context, w Want) Status {
 		}
 		return st
 	}
-	if w.Offline != nil && recipe.Repo == "" {
+	if w.Offline != nil && (recipe.Repo == "" || builds[0].Python != "") {
 		// Said, not attempted: uv would spend minutes failing to reach PyPI,
 		// and the failure would read like a network fault rather than the
 		// design of an offline site.

internal/agent/provision/python.go

@@ -109,6 +109,11 @@ func (in *Installer) installPython(ctx context.Context, r Recipe, w Want, b Buil
 		}
 		reqs = append(reqs, more...)
 	}
+	if len(b.Assets) > 0 {
+		if err := in.unpackAssets(ctx, r, w, b, tmp, st); err != nil {
+			return err
+		}
+	}
 	if b.Wheel != nil {
 		// The engine from its GitHub release, pinned to the digest GitHub
 		// publishes for the file: the same trust as an archive, and uv
@@ -372,6 +377,48 @@ func (in *Installer) unpackSource(ctx context.Context, w Want, b Build, env stri
 	return requirementsIn(string(f), b.IndexPackages)
 }
 
+// unpackAssets fetches a Python engine's release binaries, each checked
+// against the digest GitHub publishes for it, and unpacks them into the
+// environment's AssetDir.
+func (in *Installer) unpackAssets(ctx context.Context, r Recipe, w Want, b Build, env string, st *Status) error {
+	tag := b.AssetTag
+	if tag == "" {
+		tag = w.Version
+	}
+	in.progress(st, "asking GitHub for the checksums of %s %s", r.Repo, tag)
+	sums, err := in.checksums(ctx, r.Repo, tag)
+	if err != nil {
+		return err
+	}
+	dst := filepath.Join(env, b.AssetDir)
+	if err := os.MkdirAll(dst, 0o755); err != nil {
+		return err
+	}
+	for _, asset := range b.Assets {
+		want, ok := sums[asset]
+		if !ok {
+			// Refused rather than installed unchecked: this is a binary that
+			// will run on every node.
+			return fmt.Errorf("%s %s publishes no checksum for %s, so it cannot be verified", r.Repo, tag, asset)
+		}
+		url := strings.TrimSuffix(in.downloadBase(), "/") + "/" + r.Repo + "/releases/download/" + tag + "/" + asset
+		archive := filepath.Join(env+".source", asset)
+		if err := os.MkdirAll(filepath.Dir(archive), 0o755); err != nil {
+			return err
+		}
+		if err := in.download(ctx, url, asset, archive, want, st); err != nil {
+			return err
+		}
+		in.progress(st, "unpacking %s", asset)
+		err := untarGz(archive, dst)
+		_ = os.RemoveAll(filepath.Dir(archive))
+		if err != nil {
+			return fmt.Errorf("unpacking %s: %w", asset, err)
+		}
+	}
+	return nil
+}
+
 // requirementsIn reads a requirements file's lines. Options (-r, -e, --index-url
 // and the like) are refused: they would fetch from somewhere other than PyPI,
 // or install what the lock never saw.

internal/agent/provision/recipe.go

@@ -76,6 +76,11 @@ type Build struct {
 	// was published, which PyPI cannot be asked for.
 	Source *Source
 	AsOf   time.Time
+	// AssetTag and AssetDir, for a Python engine that also ships a binary
+	// in its releases, are the release tag Assets are taken from, checked
+	// against the digests GitHub publishes, and the directory in the
+	// environment they are unpacked into.
+	AssetTag, AssetDir string
 	// FreeGB is the disk space the installation needs before it starts: a
 	// Python engine is several gigabytes, and one that fills the disk halfway
 	// leaves the node worse off than one that never began.
@@ -125,6 +130,7 @@ var Recipes = map[string]Recipe{
 	"vllm":     {PyPI: "vllm", Candidates: vllmCandidates},
 	"sglang":   {PyPI: "sglang", Candidates: sglangCandidates},
 	"comfyui":  {Candidates: comfyuiCandidates},
+	"strata":   {Repo: StrataRepo, Candidates: strataCandidates},
 	"uv":       {Repo: "astral-sh/uv", Candidates: uvCandidates},
 }
 
@@ -172,6 +178,9 @@ func Builds(engine, version, arch string) ([]Build, error) {
 			continue
 		}
 		for _, b := range bs {
+			if b.Python != "" {
+				return nil, fmt.Errorf("%s is a Python environment from PyPI, which an offline bundle does not carry", engine)
+			}
 			if !seen[b.Variant] {
 				seen[b.Variant] = true
 				out = append(out, b)
@@ -672,3 +681,80 @@ print("comfyui version", "` + version + `", "on torch", torch.__version__)`},
 		{cu126, gpuNeeds{minDriverForCUDA12, 5.0}, "ComfyUI " + version + " on CUDA 12.6's torch"},
 	}, h)
 }
+
+// --- Strata ---
+
+// StrataRepo is where Strata is published: its source, and in each release
+// the prebuilt engine for each CUDA.
+const StrataRepo = "Niko1221/Strata"
+
+// strataReleases are the Strata versions this agent installs: the source
+// archive's digest, taken when it was added here and confirmed from a second
+// machine, when the version was published, and the engine its release
+// carries, whose files GitHub publishes digests for.
+var strataReleases = map[string]struct {
+	sha256    string
+	published string
+	engine    string
+}{
+	"0.2.7": {"41ae861478f5e5d9506864f97dfb5dd7cfe6b83193300e2b0d4f04c0380e3af9", "2026-10-27T14:54:42Z", "0.1.52"},
+}
+
+// strataCandidates are the Strata builds for a machine: its source, in a
+// Python environment, with the prebuilt engine for the node's driver and
+// cards — built for CUDA 13, Turing to Blackwell; or for CUDA 12.9, Pascal
+// to Ada, which is what a V100, or a driver older than 580, gets.
+func strataCandidates(version, variant string, h discovery.Host) ([]Build, error) {
+	if _, err := gpuArch("Strata", h); err != nil {
+		return nil, err
+	}
+	if h.Arch != "amd64" {
+		return nil, NotHere{"Strata publishes its Linux CUDA engine for x86-64 only"}
+	}
+	rel, ok := strataReleases[version]
+	if !ok {
+		known := make([]string, 0, len(strataReleases))
+		for v := range strataReleases {
+			known = append(known, v)
+		}
+		slices.Sort(known)
+		return nil, fmt.Errorf("this agent knows the source digest of Strata %s, not of %s; "+
+			"install that one by hand and name it in agent.yaml (engines.strata.dir)",
+			strings.Join(known, ", "), version)
+	}
+	published, err := time.Parse(time.RFC3339, rel.published)
+	if err != nil {
+		return nil, err
+	}
+	build := func(cuda string) Build {
+		return Build{
+			Variant: cuda,
+			Python:  pythonForEngines,
+			Source: &Source{
+				URL:    "/" + StrataRepo + "/archive/refs/tags/v" + version + ".tar.gz",
+				SHA256: rel.sha256, Top: "Strata-" + version, Dir: "strata",
+				Requirements: "requirements.txt",
+			},
+			AsOf:     published.Add(24 * time.Hour),
+			Assets:   []string{"strata-engine-" + rel.engine + "-linux-x86_64-cuda-" + cuda + ".tar.gz"},
+			AssetTag: "v" + version,
+			AssetDir: "strata-engine",
+			// Strata is started as a script, so what the agent is given is
+			// the environment's interpreter; the proof is importing what its
+			// server needs, and finding its engine beside it.
+			Binary: "python",
+			VersionArgs: []string{"-c", `import os, sys, numpy, jinja2, regex
+e = os.path.dirname(os.path.dirname(os.path.abspath(sys.executable)))
+for f in ("strata/setup.py", "strata/serve/server.py", "strata-engine/strata"):
+    assert os.path.isfile(os.path.join(e, f)), f + " is missing"
+print("strata version", "` + version + `", "engine", "` + rel.engine + `")`},
+			FreeGB: 5,
+		}
+	}
+	return pick("Strata "+version, variant, []candidate{
+		{build("cuda13"), gpuNeeds{minDriverForCUDA13, 7.5}, "Strata " + version + " built for CUDA 13"},
+		// 12.9 rather than 12: its machine code is for 6.0 to 8.9 and its
+		// PTX was compiled by 12.9, so the driver must be 575.
+		{build("cuda12"), gpuNeeds{575, 6.0}, "Strata " + version + " built for CUDA 12.9"},
+	}, h)
+}

internal/agent/provision/strata_test.go

@@ -0,0 +1,118 @@
+package provision
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"gitea.mynodes.xyz/ilker/llmgw/internal/agent/discovery"
+)
+
+// strataRelease is Strata's source archive as GitHub makes one, and its
+// release's engine for CUDA 13, with the recipe pointed at the archive.
+func strataRelease(t *testing.T, rel *release) {
+	t.Helper()
+	archive := tarGz(t, map[string]string{
+		"Strata-0.2.7/setup.py":         "print('setup')\n",
+		"Strata-0.2.7/serve/server.py":  "print('server')\n",
+		"Strata-0.2.7/requirements.txt": "numpy==2.5.3; python_version >= \"3.12\"\njinja2==3.1.6\n",
+	}, nil)
+	rel.assets["v0.2.7.tar.gz"] = archive
+	sum := sha256.Sum256(archive)
+	engine := tarGz(t, map[string]string{"strata": "\x7fELF", "libcudart.so.13": "lib"}, nil)
+	name := "strata-engine-0.1.52-linux-x86_64-cuda-cuda13.tar.gz"
+	rel.assets[name] = engine
+	esum := sha256.Sum256(engine)
+	rel.digests[name] = "sha256:" + hex.EncodeToString(esum[:])
+
+	old := strataReleases
+	strataReleases = map[string]struct {
+		sha256    string
+		published string
+		engine    string
+	}{"0.2.7": {hex.EncodeToString(sum[:]), "2026-10-27T14:54:42Z", "0.1.52"}}
+	t.Cleanup(func() { strataReleases = old })
+}
+
+func TestStrataIsInstalledWithItsEngineBesideItsCode(t *testing.T) {
+	sys := &fakePython{fakeSystem: fakeSystem{installed: map[string]bool{}}}
+	in, rel := pythonInstaller(t, turingNode, sys)
+	strataRelease(t, rel)
+
+	got := in.Converge(t.Context(), []Want{{Engine: "strata", Version: "0.2.7"}})
+	if got[0].State != Ready {
+		t.Fatalf("not ready: %+v", got[0])
+	}
+	dir := filepath.Join(in.Root, "strata", "0.2.7-cuda13")
+	if got[0].Binary != filepath.Join(dir, "bin", "python") {
+		t.Errorf("binary %s, want the environment's interpreter", got[0].Binary)
+	}
+	for _, f := range []string{"strata/setup.py", "strata/serve/server.py", "strata-engine/strata", "strata-engine/libcudart.so.13"} {
+		if _, err := os.Stat(filepath.Join(dir, f)); err != nil {
+			t.Errorf("%s is not in the environment: %v", f, err)
+		}
+	}
+	all := strings.Join(sys.commands, "\n")
+	for _, want := range []string{"--exclude-newer 2026-10-28T14:54:42Z", "jinja2==3.1.6"} {
+		if !strings.Contains(all, want) {
+			t.Errorf("the installation never used %q:\n%s", want, all)
+		}
+	}
+}
+
+func TestStrataRefusesAnEngineWhoseDigestDoesNotMatch(t *testing.T) {
+	sys := &fakePython{fakeSystem: fakeSystem{installed: map[string]bool{}}}
+	in, rel := pythonInstaller(t, turingNode, sys)
+	strataRelease(t, rel)
+	rel.digests["strata-engine-0.1.52-linux-x86_64-cuda-cuda13.tar.gz"] = "sha256:" + strings.Repeat("0", 64)
+
+	got := in.Converge(t.Context(), []Want{{Engine: "strata", Version: "0.2.7", Variant: "cuda13"}})
+	if got[0].State != Failed || !strings.Contains(got[0].Detail, "does not match the checksum") {
+		t.Fatalf("an engine that does not match its digest was not refused: %+v", got[0])
+	}
+}
+
+func TestStrataPicksTheEngineForTheCardsAndDriver(t *testing.T) {
+	for _, tc := range []struct {
+		name string
+		host discovery.Host
+		want []string
+		err  string
+	}{
+		{"turing on 580", turingNode, []string{"cuda13", "cuda12"}, ""},
+		{"volta", discovery.Host{Arch: "amd64", NVIDIADriver: "580.82.07", MinComputeCapability: 7.0}, []string{"cuda12"}, ""},
+		{"driver 575", discovery.Host{Arch: "amd64", NVIDIADriver: "575.51.03", MinComputeCapability: 8.6}, []string{"cuda12"}, ""},
+		{"driver 570", discovery.Host{Arch: "amd64", NVIDIADriver: "570.86.10", MinComputeCapability: 8.6}, nil, "575"},
+		{"no gpu", discovery.Host{Arch: "amd64"}, nil, "no"},
+		{"arm", discovery.Host{Arch: "arm64", NVIDIADriver: "580.82.07", MinComputeCapability: 9.0}, nil, "x86-64"},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			bs, err := strataCandidates("0.2.7", "", tc.host)
+			if tc.err != "" {
+				if err == nil || !strings.Contains(err.Error(), tc.err) {
+					t.Fatalf("got %v, %v; want an error saying %q", bs, err, tc.err)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatal(err)
+			}
+			var got []string
+			for _, b := range bs {
+				got = append(got, b.Variant)
+			}
+			if strings.Join(got, ",") != strings.Join(tc.want, ",") {
+				t.Errorf("builds %v, want %v", got, tc.want)
+			}
+		})
+	}
+}
+
+func TestStrataIsNotInAnOfflineBundle(t *testing.T) {
+	if _, err := Builds("strata", "0.2.7", "amd64"); err == nil {
+		t.Fatal("an offline bundle would carry Strata, which needs PyPI")
+	}
+}

internal/agent/weights/fetch.go

@@ -90,14 +90,20 @@ var ErrNotLocal = errors.New("the weights are not on this node and the source sa
 // already cached, downloaded, or downloaded by another shard this one waited
 // for — since that is most of the difference between a replica that starts
 // in a minute and one that takes an hour.
-func (f *Fetcher) Ensure(ctx context.Context, source string) (path string, err error) {
+//
+// A source in a pickle format, or a set with such a file in it, is refused
+// before anything is fetched unless allowPickle; see Pickled.
+func (f *Fetcher) Ensure(ctx context.Context, source string, allowPickle bool) (path string, err error) {
 	ctx, span := obs.Tracer(tracerName).Start(ctx, "fetch weights",
 		trace.WithAttributes(attribute.String("llmgw.weights.source", redact(source))))
 	defer func() {
 		span.SetAttributes(attribute.String("llmgw.weights.path", path))
 		obs.End(span, err)
 	}()
-	return f.ensure(ctx, source)
+	if !allowPickle && !IsSet(source) && Pickled(source) {
+		return "", refusePickle(source)
+	}
+	return f.ensure(ctx, source, allowPickle)
 }
 
 // tracerName names the spans this package opens.
@@ -119,7 +125,7 @@ func redact(source string) string {
 	return u.String()
 }
 
-func (f *Fetcher) ensure(ctx context.Context, source string) (string, error) {
+func (f *Fetcher) ensure(ctx context.Context, source string, allowPickle bool) (string, error) {
 	if source == "" {
 		return "", fmt.Errorf("no source given")
 	}
@@ -127,7 +133,7 @@ func (f *Fetcher) ensure(ctx context.Context, source string) (string, error) {
 	// file, and a set has to be handled before any of it.
 	if IsSet(source) {
 		outcome(ctx, "set")
-		return f.ensureSet(ctx, source)
+		return f.ensureSet(ctx, source, allowPickle)
 	}
 	if !isRemote(source) {
 		if !isPath(source) {

internal/agent/weights/fetch_test.go

@@ -37,7 +37,7 @@ func TestALocalPathIsUsedNeverCopied(t *testing.T) {
 	}
 	f := &Fetcher{CacheDir: t.TempDir()}
 
-	got, err := f.Ensure(t.Context(), path)
+	got, err := f.Ensure(t.Context(), path, false)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -53,7 +53,7 @@ func TestAMissingLocalPathSaysWhatIsWrong(t *testing.T) {
 	// The commonest setup mistake, and "no such file" from an engine twenty
 	// minutes into a load is a poor way to learn it.
 	f := &Fetcher{CacheDir: t.TempDir()}
-	_, err := f.Ensure(t.Context(), filepath.Join(t.TempDir(), "absent.gguf"))
+	_, err := f.Ensure(t.Context(), filepath.Join(t.TempDir(), "absent.gguf"), false)
 	if !errors.Is(err, ErrNotLocal) {
 		t.Fatalf("err = %v, want ErrNotLocal", err)
 	}
@@ -67,7 +67,7 @@ func TestARemoteFileIsFetchedOnceAndReused(t *testing.T) {
 	srv := serveBytes(t, []byte("the weights"), &hits)
 	f := &Fetcher{CacheDir: t.TempDir()}
 
-	first, err := f.Ensure(t.Context(), srv.URL+"/model.gguf")
+	first, err := f.Ensure(t.Context(), srv.URL+"/model.gguf", false)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -76,7 +76,7 @@ func TestARemoteFileIsFetchedOnceAndReused(t *testing.T) {
 		t.Fatalf("fetched content = %q, err %v", body, err)
 	}
 
-	second, err := f.Ensure(t.Context(), srv.URL+"/model.gguf")
+	second, err := f.Ensure(t.Context(), srv.URL+"/model.gguf", false)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -103,7 +103,7 @@ func TestConcurrentRequestsProduceOneDownload(t *testing.T) {
 		wg.Add(1)
 		go func() {
 			defer wg.Done()
-			paths[i], errs[i] = f.Ensure(t.Context(), srv.URL+"/model.gguf")
+			paths[i], errs[i] = f.Ensure(t.Context(), srv.URL+"/model.gguf", false)
 		}()
 	}
 	wg.Wait()
@@ -139,7 +139,7 @@ func TestAPartialDownloadIsNeverVisibleAsTheModel(t *testing.T) {
 
 	cache := t.TempDir()
 	f := &Fetcher{CacheDir: cache}
-	if _, err := f.Ensure(t.Context(), srv.URL+"/model.gguf"); err == nil {
+	if _, err := f.Ensure(t.Context(), srv.URL+"/model.gguf", false); err == nil {
 		t.Fatal("a truncated download was reported as success")
 	}
 	entries, _ := os.ReadDir(cache)
@@ -169,7 +169,7 @@ func TestAnInterruptedDownloadResumes(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	got, err := f.Ensure(t.Context(), srv.URL+"/model.gguf")
+	got, err := f.Ensure(t.Context(), srv.URL+"/model.gguf", false)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -232,7 +232,7 @@ func TestProgressIsReportedWithAPercentage(t *testing.T) {
 		CacheDir:   t.TempDir(),
 		OnProgress: func(p Progress) { seen = append(seen, p) },
 	}
-	if _, err := f.Ensure(t.Context(), srv.URL+"/model.gguf"); err != nil {
+	if _, err := f.Ensure(t.Context(), srv.URL+"/model.gguf", false); err != nil {
 		t.Fatal(err)
 	}
 	if len(seen) == 0 {
@@ -254,7 +254,7 @@ func TestPercentIsUnknownRatherThanZeroWithoutATotal(t *testing.T) {
 
 func TestNoCacheDirectoryIsAClearRefusal(t *testing.T) {
 	f := &Fetcher{}
-	_, err := f.Ensure(t.Context(), "https://example.invalid/model.gguf")
+	_, err := f.Ensure(t.Context(), "https://example.invalid/model.gguf", false)
 	if err == nil || !strings.Contains(err.Error(), "cache directory") {
 		t.Errorf("err = %v, want a complaint about the cache directory", err)
 	}
@@ -269,7 +269,7 @@ func TestABareModelIdIsLeftToTheEngine(t *testing.T) {
 		"Qwen/Qwen2.5-7B-Instruct",
 		"meta-llama/Llama-3.3-70B-Instruct",
 	} {
-		got, err := f.Ensure(t.Context(), id)
+		got, err := f.Ensure(t.Context(), id, false)
 		if err != nil {
 			t.Errorf("%s was refused: %v", id, err)
 		}
@@ -283,7 +283,7 @@ func TestAnAbsolutePathThatIsMissingIsStillAnError(t *testing.T) {
 	// The other half: an operator writing a path means a file, and a missing
 	// one should be said plainly rather than handed to an engine.
 	f := &Fetcher{CacheDir: t.TempDir()}
-	if _, err := f.Ensure(t.Context(), "/models/definitely-not-here.gguf"); !errors.Is(err, ErrNotLocal) {
+	if _, err := f.Ensure(t.Context(), "/models/definitely-not-here.gguf", false); !errors.Is(err, ErrNotLocal) {
 		t.Errorf("err = %v, want ErrNotLocal", err)
 	}
 }

internal/agent/weights/pickle.go

@@ -0,0 +1,43 @@
+package weights
+
+import (
+	"errors"
+	"fmt"
+	"path"
+	"strings"
+)
+
+// Weights in Python's pickle format are a program as much as a file: loading
+// one runs whatever it says, as the engine's user, on the node. That is how
+// the malicious models found on Hugging Face work, and nothing in the file's
+// name or size says so. safetensors and GGUF are only tensors and are read,
+// never run, so those are what a node fetches unless the catalog says
+// otherwise for a model, with allow_pickle, on purpose.
+
+// pickleExts are the extensions of PyTorch's pickle formats, and of pickle
+// itself. A .bin beside safetensors is the same model in pickle form.
+var pickleExts = []string{".bin", ".pt", ".pt2", ".pth", ".ckpt", ".pkl", ".pickle"}
+
+// Pickled reports whether a file name or URL is in a pickle format.
+func Pickled(name string) bool {
+	// A query or a fragment is not the file's name, on a URL or an hf:
+	// source alike: model.bin?download=true is still model.bin.
+	if i := strings.IndexAny(name, "?#"); i >= 0 {
+		name = name[:i]
+	}
+	ext := strings.ToLower(path.Ext(name))
+	for _, e := range pickleExts {
+		if ext == e {
+			return true
+		}
+	}
+	return false
+}
+
+// ErrPickle is a source refused for being in a pickle format.
+var ErrPickle = errors.New("weights in a pickle format run code when they are loaded")
+
+func refusePickle(name string) error {
+	return fmt.Errorf("%w: %s is refused; use the model's safetensors or GGUF, or set allow_pickle: true "+
+		"on the model if it was checked and has no other form", ErrPickle, redact(name))
+}

internal/agent/weights/pickle_test.go

@@ -0,0 +1,66 @@
+package weights
+
+import (
+	"errors"
+	"sync/atomic"
+	"testing"
+)
+
+func TestPickleFormatsAreKnownByTheirNames(t *testing.T) {
+	for name, want := range map[string]bool{
+		"pytorch_model.bin": true, "model.ckpt": true, "x.PT": true, "w.pth": true, "a.pkl": true,
+		"https://h/r/resolve/main/model.ckpt?download=true": true,
+		"hf:o/r/model.bin?download=true":                    true, "hf:o/r/model.ckpt#x": true, "m.pt2": true,
+		"model.safetensors": false, "m-Q4_K_M.gguf": false, "Qwen/Qwen2.5-7B-Instruct": false,
+		"config.json": false, "": false,
+	} {
+		if got := Pickled(name); got != want {
+			t.Errorf("Pickled(%q) = %v, want %v", name, got, want)
+		}
+	}
+}
+
+func TestAPickleFileIsRefusedBeforeItIsFetched(t *testing.T) {
+	var hits atomic.Int64
+	srv := serveBytes(t, []byte("not really a pickle"), &hits)
+	f := &Fetcher{CacheDir: t.TempDir()}
+
+	_, err := f.Ensure(t.Context(), srv.URL+"/model.ckpt", false)
+	if !errors.Is(err, ErrPickle) {
+		t.Fatalf("a .ckpt was not refused: %v", err)
+	}
+	if hits.Load() != 0 {
+		t.Errorf("the refused file was fetched %d times", hits.Load())
+	}
+	if _, err := f.Ensure(t.Context(), srv.URL+"/model.ckpt", true); err != nil {
+		t.Errorf("allowed, the .ckpt was still refused: %v", err)
+	}
+}
+
+func TestASetWithAPickleFileIsRefusedWhole(t *testing.T) {
+	f, srv, hits := mirrorOf(t, map[string]string{
+		"pytorch_model.bin": "pickle",
+		"config.json":       "{}",
+	})
+	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false); !errors.Is(err, ErrPickle) {
+		t.Fatalf("a set holding pytorch_model.bin was not refused: %v", err)
+	}
+	if *hits != 0 {
+		t.Errorf("%d files of the refused set were fetched", *hits)
+	}
+	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/", true); err != nil {
+		t.Errorf("allowed, the set was still refused: %v", err)
+	}
+}
+
+func TestASetAlreadyHereIsLookedAtAgain(t *testing.T) {
+	// Fetched once for a model that allowed pickles, or before they were
+	// refused: another asking for it without leave is refused all the same.
+	f, srv, _ := mirrorOf(t, map[string]string{"pytorch_model.bin": "pickle", "config.json": "{}"})
+	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/", true); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false); !errors.Is(err, ErrPickle) {
+		t.Errorf("a set already on the node was not looked at: %v", err)
+	}
+}

internal/agent/weights/set.go

@@ -46,7 +46,7 @@ type manifest struct {
 // The whole set or nothing. A directory with three of four shards in it loads
 // as a broken model rather than failing to load, so a partial set is left
 // unfinished under its temporary name and the next attempt starts again.
-func (f *Fetcher) ensureSet(ctx context.Context, source string) (string, error) {
+func (f *Fetcher) ensureSet(ctx context.Context, source string, allowPickle bool) (string, error) {
 	dir, err := f.cachePath(strings.TrimSuffix(source, "/"))
 	if err != nil {
 		return "", err
@@ -56,6 +56,16 @@ func (f *Fetcher) ensureSet(ctx context.Context, source string) (string, error)
 	// exactly the failure the marker exists to prevent.
 	done := filepath.Join(dir, ".complete")
 	if _, err := os.Stat(done); err == nil {
+		// Already here, maybe fetched for a model that allowed pickles, or
+		// before they were refused: looked at again, not trusted.
+		if !allowPickle {
+			entries, _ := os.ReadDir(dir)
+			for _, e := range entries {
+				if Pickled(e.Name()) {
+					return "", refusePickle(source + e.Name())
+				}
+			}
+		}
 		return dir, nil
 	}
 
@@ -66,6 +76,15 @@ func (f *Fetcher) ensureSet(ctx context.Context, source string) (string, error)
 	if len(m.Files) == 0 {
 		return "", fmt.Errorf("%s holds no files", source)
 	}
+	// Refused whole rather than skipped: an engine given the directory
+	// loads the pickle when there is no safetensors beside it.
+	if !allowPickle {
+		for _, file := range m.Files {
+			if Pickled(file.Name) {
+				return "", refusePickle(source + file.Name)
+			}
+		}
+	}
 
 	if err := os.MkdirAll(dir, 0o755); err != nil {
 		return "", err

internal/agent/weights/set_test.go

@@ -53,7 +53,7 @@ func TestASetIsFetchedIntoOneDirectory(t *testing.T) {
 		"config.json":                      "{}",
 	})
 
-	got, err := f.Ensure(t.Context(), srv.URL+"/m/v/")
+	got, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -74,11 +74,11 @@ func TestASetIsFetchedIntoOneDirectory(t *testing.T) {
 func TestAFinishedSetIsNotFetchedAgain(t *testing.T) {
 	f, srv, hits := mirrorOf(t, map[string]string{"a.gguf": "aaaa", "b.gguf": "bb"})
 
-	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/"); err != nil {
+	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false); err != nil {
 		t.Fatal(err)
 	}
 	after := *hits
-	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/"); err != nil {
+	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false); err != nil {
 		t.Fatal(err)
 	}
 	if *hits != after {
@@ -100,7 +100,7 @@ func TestAHalfFetchedSetIsNotTreatedAsFinished(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	got, err := f.Ensure(t.Context(), srv.URL+"/m/v/")
+	got, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -117,7 +117,7 @@ func TestAFileAlreadyThereAtTheRightSizeIsNotRefetched(t *testing.T) {
 	_ = os.MkdirAll(dir, 0o755)
 	_ = os.WriteFile(filepath.Join(dir, "a.gguf"), []byte("aaaa"), 0o644)
 
-	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/"); err != nil {
+	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false); err != nil {
 		t.Fatal(err)
 	}
 	// Resuming a set means not paying for the shards that already landed.
@@ -133,7 +133,7 @@ func TestAFilenameThatEscapesTheDirectoryIsRefused(t *testing.T) {
 	defer srv.Close()
 	f := &Fetcher{CacheDir: t.TempDir()}
 
-	_, err := f.Ensure(t.Context(), srv.URL+"/m/v/")
+	_, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false)
 	// The mirror is a place an operator copies files into by hand, so what it
 	// reports is not automatically safe to join onto a path.
 	if err == nil || !strings.Contains(err.Error(), "offered a file named") {
@@ -148,7 +148,7 @@ func TestADirectoryListingIsNotAManifest(t *testing.T) {
 	defer srv.Close()
 	f := &Fetcher{CacheDir: t.TempDir()}
 
-	_, err := f.Ensure(t.Context(), srv.URL+"/m/v/")
+	_, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false)
 	// "invalid character '<'" sends somebody looking at the wrong thing.
 	if err == nil || !strings.Contains(err.Error(), "has to be a mirror version") {
 		t.Errorf("the reason should name the cause, got %v", err)
@@ -182,12 +182,12 @@ func TestASetFileThatDoesNotMatchTheMirrorsChecksumIsRefused(t *testing.T) {
 	}))
 	t.Cleanup(srv.Close)
 	f := &Fetcher{CacheDir: t.TempDir()}
-	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/"); err == nil || !strings.Contains(err.Error(), "sha256") {
+	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false); err == nil || !strings.Contains(err.Error(), "sha256") {
 		t.Fatalf("a damaged shard was accepted: %v", err)
 	}
 	served = good
 	// The size differs, so it is fetched again; this time it matches.
-	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/"); err != nil {
+	if _, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false); err != nil {
 		t.Fatalf("the shard as mirrored: %v", err)
 	}
 }
@@ -198,7 +198,7 @@ func TestASetStillDownloadingOnTheMirrorSaysSo(t *testing.T) {
 	}))
 	t.Cleanup(srv.Close)
 	f := &Fetcher{CacheDir: t.TempDir()}
-	_, err := f.Ensure(t.Context(), srv.URL+"/m/v/")
+	_, err := f.Ensure(t.Context(), srv.URL+"/m/v/", false)
 	if err == nil || !strings.Contains(err.Error(), "not complete on the mirror yet") ||
 		!strings.Contains(err.Error(), "model-00002-of-00002") {
 		t.Fatalf("got %v", err)

internal/agent/weights/trace_test.go

@@ -13,7 +13,7 @@ func TestAFetchSaysInItsTraceWhetherItDownloadedAndHowMuch(t *testing.T) {
 	f := &Fetcher{CacheDir: t.TempDir()}
 	src := strings.Replace(srv.URL, "http://", "http://user:secret@", 1) + "/m/model.gguf?token=hunter2"
 
-	if _, err := f.Ensure(t.Context(), src); err != nil {
+	if _, err := f.Ensure(t.Context(), src, false); err != nil {
 		t.Fatal(err)
 	}
 	fetch := rec.One(t, [16]byte{}, "fetch weights")
@@ -30,7 +30,7 @@ func TestAFetchSaysInItsTraceWhetherItDownloadedAndHowMuch(t *testing.T) {
 	}
 
 	// The second start on this node finds it here.
-	if _, err := f.Ensure(t.Context(), src); err != nil {
+	if _, err := f.Ensure(t.Context(), src, false); err != nil {
 		t.Fatal(err)
 	}
 	again := rec.Named("fetch weights")

internal/cli/mirror.go

@@ -51,6 +51,7 @@ func newMirrorCmd() *cobra.Command {
 		newMirrorRemoveCmd(&dir),
 		newMirrorServeCmd(&dir),
 		newMirrorInitCmd(&dir),
+		newMirrorAddPeerCmd(&dir),
 	)
 	return c
 }
@@ -255,6 +256,7 @@ func newMirrorServeCmd(dir *string) *cobra.Command {
 	var tlsc tlsconf.Config
 	var allow, clients []string
 	var allowPrivate bool
+	var peers []string
 
 	c := &cobra.Command{
 		Use:   "serve",
@@ -317,6 +319,21 @@ func newMirrorServeCmd(dir *string) *cobra.Command {
 				ReadTimeout: time.Minute,
 			})
 
+			// Another mirror kept in step with this one, each copying what
+			// the other has: the cluster's models while one machine is down.
+			if len(peers) > 0 {
+				p := &mirror.Peers{M: mgr, Log: log}
+				for _, addr := range peers {
+					c, err := mirror.NewClient(addr, tlsc, 30*time.Second)
+					if err != nil {
+						return fmt.Errorf("--peer %s: %w", addr, err)
+					}
+					p.Clients = append(p.Clients, c)
+				}
+				go p.Run(cmd.Context())
+				fmt.Fprintf(out, "kept in step with %s, every %s\n", strings.Join(peers, ", "), mirror.PeerInterval)
+			}
+
 			errs := make(chan error, 2)
 			go func() { errs <- fmt.Errorf("serving weights: %w", files.ListenAndServe()) }()
 			fmt.Fprintf(out, "serving %s on %s\n", *dir, listen)
@@ -343,6 +360,9 @@ func newMirrorServeCmd(dir *string) *cobra.Command {
 		},
 	}
 	c.Flags().StringVar(&listen, "listen", ":8081", "address to serve the weights on")
+	c.Flags().StringArrayVar(&peers, "peer", nil,
+		"another mirror's management API to keep in step with, each copying what the other has; repeatable. "+
+			"Give each the other's address in --api-allow (or its certificate in --api-client)")
 	c.Flags().StringVar(&apiListen, "api-listen", ":8082", "address to serve the management API on")
 	c.Flags().StringVar(&base, "base", "http://localhost:8081",
 		"the address nodes use to reach this mirror, as it should appear in catalog sources")

internal/cli/mirror_peer.go

@@ -0,0 +1,209 @@
+package cli
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"net"
+	"os"
+	"regexp"
+	"slices"
+	"strings"
+
+	"github.com/spf13/cobra"
+
+	"gitea.mynodes.xyz/ilker/llmgw/internal/onboard"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/settings"
+)
+
+// A second mirror, so models can still be added and fetched while the first
+// machine is down. Set up from the machine of the first, which it is kept in
+// step with: the first is told of it, it of the first, and the control
+// planes of both.
+
+func newMirrorAddPeerCmd(dir *string) *cobra.Command {
+	var controlPlanes []string
+	var base string
+	cmd := &cobra.Command{
+		Use:   "add-peer [user@]host",
+		Short: "Set up a second mirror on another machine, kept in step with this one",
+		Long: `Run on the machine of the cluster's mirror, as root. It logs in to HOST with
+ssh, installs llmgw there if it is not, and sets up a mirror that keeps in
+step with this one: each copies what the other holds, checksum checked, and
+a model removed from one is removed from the other. This mirror is told of
+it, and the cluster's settings name it (mirror.peers), so while either
+machine is down the control planes use the other and nodes fetch from it.
+
+HOST needs room for every model this mirror holds: it copies them all.`,
+		Example: "  sudo llmgw mirror add-peer ubuntu@10.0.0.12 --control-plane 10.0.0.10 --control-plane 10.0.0.12",
+		Args:    cobra.ExactArgs(1),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if os.Geteuid() != 0 {
+				return errors.New("add-peer changes this mirror's service and the cluster's settings: run it as root")
+			}
+			user, host, ok := strings.Cut(args[0], "@")
+			if !ok {
+				user, host = os.Getenv("SUDO_USER"), args[0]
+			}
+			return addMirrorPeer(cmd.Context(), cmd.OutOrStdout(), &onboard.SSH{Addr: host, User: user}, host,
+				*dir, base, controlPlanes, "/etc/llmgw/server.yaml")
+		},
+	}
+	cmd.Flags().StringArrayVar(&controlPlanes, "control-plane", nil,
+		"address of a control plane that manages the mirrors; repeatable (default: this machine)")
+	cmd.Flags().StringVar(&base, "base", "", "the address nodes fetch from on HOST (default: http://HOST:8081)")
+	return cmd
+}
+
+// addMirrorPeer sets up a mirror on a machine reached by remote, kept in
+// step with the one on this machine.
+func addMirrorPeer(ctx context.Context, out io.Writer, remote onboard.Remote, host, dir, base string,
+	controlPlanes []string, configPath string,
+) error {
+	ip, err := ipv4Of(host)
+	if err != nil {
+		return err
+	}
+	here := primaryAddr()
+	if len(controlPlanes) == 0 {
+		controlPlanes = []string{here}
+	}
+	if base == "" {
+		base = "http://" + net.JoinHostPort(ip, "8081")
+	}
+	thisAPI := "http://" + net.JoinHostPort(here, "8082")
+	peerAPI := "http://" + net.JoinHostPort(ip, "8082")
+
+	// llmgw there, the same as here, when it has none.
+	if _, err := remote.Run(ctx, "command -v llmgw >/dev/null", nil); err != nil {
+		exe, err := os.Executable()
+		if err != nil {
+			return err
+		}
+		bin, err := os.ReadFile(exe) //nolint:gosec // this program
+		if err != nil {
+			return err
+		}
+		fmt.Fprintf(out, "  ..    llmgw to %s\n", host)
+		if _, err := remote.Run(ctx, "sudo install -m 755 /dev/stdin /usr/local/bin/llmgw", bin); err != nil {
+			return fmt.Errorf("installing llmgw on %s: %w", host, err)
+		}
+	}
+
+	// The mirror there: its management API for the control planes and this
+	// mirror, which reads it; kept in step with this one.
+	allow := append([]string{}, controlPlanes...)
+	if !slices.Contains(allow, here) {
+		allow = append(allow, here)
+	}
+	script := "sudo llmgw mirror init --dir " + shellQuote(dir) + " --base " + shellQuote(base) + " --peer " + shellQuote(thisAPI)
+	for _, a := range allow {
+		script += " --control-plane " + shellQuote(a)
+	}
+	fmt.Fprintf(out, "  ..    the mirror on %s\n", host)
+	if b, err := remote.Run(ctx, script, nil); err != nil {
+		return fmt.Errorf("the mirror on %s: %w\n%s", host, err, b)
+	}
+
+	// This mirror told of it, and to let it read this one's API.
+	changed, err := addToMirrorUnit(mirrorUnitPath, peerAPI, ip)
+	if err != nil {
+		return err
+	}
+	if changed {
+		in := &serverInit{out: out}
+		if err := in.systemctl(ctx, "daemon-reload"); err != nil {
+			return err
+		}
+		if err := in.systemctl(ctx, "restart", "llmgw-mirror"); err != nil {
+			return err
+		}
+	}
+	fmt.Fprintf(out, "  ok    this mirror keeps in step with %s\n", peerAPI)
+
+	// The control planes told of both.
+	if err := addMirrorPeerSetting(configPath, peerAPI); err != nil {
+		return err
+	}
+	fmt.Fprintf(out, "  ok    mirror.peers names %s; the control planes use it while this one is down\n", peerAPI)
+	return nil
+}
+
+// addToMirrorUnit adds a peer and the address allowed to read this mirror's
+// API to its systemd unit, and says whether it changed.
+func addToMirrorUnit(path, peer, allow string) (bool, error) {
+	b, err := os.ReadFile(path) //nolint:gosec // the mirror's unit
+	if err != nil {
+		return false, fmt.Errorf("this machine has no mirror to keep in step (%s): run add-peer on the mirror's machine", path)
+	}
+	unit := string(b)
+	re := regexp.MustCompile(`(?m)^ExecStart=.*$`)
+	line := re.FindString(unit)
+	if line == "" {
+		return false, fmt.Errorf("%s has no ExecStart", path)
+	}
+	next := line
+	if !strings.Contains(next, "--peer "+peer) {
+		next += " --peer " + peer
+	}
+	if !strings.Contains(next, "--api-allow "+allow+" ") && !strings.HasSuffix(next, "--api-allow "+allow) &&
+		strings.Contains(next, "--api-listen :") {
+		next += " --api-allow " + allow
+	}
+	if next == line {
+		return false, nil
+	}
+	return true, os.WriteFile(path, []byte(strings.Replace(unit, line, next, 1)), 0o644)
+}
+
+// addMirrorPeerSetting names a peer in mirror.peers: in the cluster's
+// settings when the store holds them, else in server.yaml.
+func addMirrorPeerSetting(path, peer string) error {
+	add := func(doc *settings.Doc) error {
+		var peers []string
+		if n := doc.Node([]string{"mirror", "peers"}); n != nil {
+			_ = n.Decode(&peers)
+		}
+		if slices.Contains(peers, peer) {
+			return nil
+		}
+		list := "[" + strings.Join(append(peers, peer), ", ") + "]"
+		return doc.SetYAML([]string{"mirror", "peers"}, list)
+	}
+	changed, err := changeStoredSettings(path, "a second mirror (llmgw mirror add-peer)", add)
+	if err != nil || changed {
+		return err
+	}
+	doc, err := settings.Open(path)
+	if err != nil {
+		return err
+	}
+	if err := add(doc); err != nil {
+		return err
+	}
+	_, err = doc.Save(nil)
+	return err
+}
+
+// ensureSecondMirror sets up the mirror kept in step with this machine's on
+// the store machine chosen, reached as the store's setup reached it.
+func (in *serverInit) ensureSecondMirror(ctx context.Context) error {
+	if !in.standbys {
+		return errors.New("it needs --standbys: without them this mirror's API answers this machine alone, which the second cannot read")
+	}
+	i := in.secondMirror - 1
+	if i < 0 || i >= len(in.patroniNodes) {
+		return fmt.Errorf("there is no database machine %d to put it on", in.secondMirror)
+	}
+	if in.noMirror || in.mirrorURL != "" {
+		return errors.New("it keeps in step with a mirror on this machine, and there is none")
+	}
+	n := in.patroniNodes[i]
+	fmt.Fprintf(in.out, "  ..    a second mirror on %s\n", n.Name)
+	if err := addMirrorPeer(ctx, in.out, n.Remote, n.IP, in.mirrorRoot(), "", in.mirrorAllow(), in.configPath()); err != nil {
+		return err
+	}
+	in.done("a second mirror on %s (%s), kept in step with this one", n.Name, n.IP)
+	return nil
+}

internal/cli/mirror_peer_test.go

@@ -0,0 +1,74 @@
+package cli
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestAPeerIsAddedToTheMirrorsUnitOnce(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "llmgw-mirror.service")
+	unit := mirrorUnitFor("/srv/models", ":8082", "http://10.0.0.10:8081", []string{"10.0.0.10"}, nil, "test")
+	if err := os.WriteFile(path, []byte(unit), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	changed, err := addToMirrorUnit(path, "http://10.0.0.12:8082", "10.0.0.12")
+	if err != nil || !changed {
+		t.Fatalf("changed %v %v", changed, err)
+	}
+	b, _ := os.ReadFile(path)
+	if !strings.Contains(string(b), "--peer http://10.0.0.12:8082") || !strings.Contains(string(b), "--api-allow 10.0.0.12") {
+		t.Errorf("unit:\n%s", b)
+	}
+	if changed, _ := addToMirrorUnit(path, "http://10.0.0.12:8082", "10.0.0.12"); changed {
+		t.Error("added twice")
+	}
+	// A mirror whose API is on loopback alone is not opened by the back door.
+	lo := mirrorUnitFor("/srv/models", "127.0.0.1:8082", "http://10.0.0.10:8081", nil, nil, "test")
+	_ = os.WriteFile(path, []byte(lo), 0o644)
+	_, _ = addToMirrorUnit(path, "http://10.0.0.12:8082", "10.0.0.12")
+	if b, _ := os.ReadFile(path); strings.Contains(string(b), "--api-allow") {
+		t.Errorf("a loopback mirror was opened:\n%s", b)
+	}
+}
+
+func TestASecondMirrorIsNamedInTheSettings(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "server.yaml")
+	if err := os.WriteFile(path, []byte("listen: \":8080\"\nmirror:\n  api: http://10.0.0.10:8082\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	for range 2 {
+		if err := addMirrorPeerSetting(path, "http://10.0.0.12:8082"); err != nil {
+			t.Fatal(err)
+		}
+	}
+	b, _ := os.ReadFile(path)
+	if strings.Count(string(b), "10.0.0.12:8082") != 1 || !strings.Contains(string(b), "peers") {
+		t.Errorf("server.yaml:\n%s", b)
+	}
+}
+
+func TestTheSetupOffersASecondMirrorOnAStoreMachine(t *testing.T) {
+	in := newDefaultServerInit()
+	f := newInstallForm(&in, "ctl", []string{"10.0.0.5"}, true)
+	f.secondMirror.Choice = 1
+	if !f.secondMirror.Show() {
+		t.Fatal("not offered with the mirror on this machine")
+	}
+	f.mirrorWhere.Choice = 1
+	if f.secondMirror.Show() {
+		t.Error("offered with the mirror elsewhere")
+	}
+	f.mirrorWhere.Choice = 0
+	in.secondMirror = 1
+	in.patroni = []string{"ubuntu@10.0.0.11", "ubuntu@10.0.0.12"}
+	if got := strings.Join(in.initFlags(), " "); !strings.Contains(got, "--second-mirror 1") {
+		t.Errorf("flags %s", got)
+	}
+	// Without standbys this mirror answers itself alone: the second could not read it.
+	in.standbys = false
+	if err := in.ensureSecondMirror(t.Context()); err == nil || !strings.Contains(err.Error(), "standbys") {
+		t.Errorf("without standbys: %v", err)
+	}
+}

internal/cli/mirrorinit.go

@@ -29,6 +29,7 @@ import (
 func newMirrorInitCmd(dir *string) *cobra.Command {
 	var (
 		controlPlanes []string
+		peers         []string
 		base          string
 		noFirewall    bool
 	)
@@ -61,7 +62,7 @@ Safe to run again: it changes only what differs.`,
 			if _, err := parseAllow(controlPlanes); err != nil {
 				return err
 			}
-			if err := mirrorInit(cmd.Context(), cmd.OutOrStdout(), *dir, base, controlPlanes); err != nil {
+			if err := mirrorInit(cmd.Context(), cmd.OutOrStdout(), *dir, base, controlPlanes, peers); err != nil {
 				return err
 			}
 			if !noFirewall {
@@ -80,6 +81,9 @@ Safe to run again: it changes only what differs.`,
 	}
 	cmd.Flags().StringArrayVar(&controlPlanes, "control-plane", nil,
 		"address of the control plane that manages this mirror; repeatable")
+	cmd.Flags().StringArrayVar(&peers, "peer", nil,
+		"another mirror's management API to keep in step with, http://host:8082; repeatable. "+
+			"Its machine must be among --control-plane, to read this one")
 	cmd.Flags().BoolVar(&noFirewall, "no-firewall", false,
 		"leave the firewall alone, instead of keeping the mirror's ports to the cluster's networks")
 	cmd.Flags().StringVar(&base, "base", "",
@@ -87,7 +91,7 @@ Safe to run again: it changes only what differs.`,
 	return cmd
 }
 
-func mirrorInit(ctx context.Context, out io.Writer, dir, base string, controlPlanes []string) error {
+func mirrorInit(ctx context.Context, out io.Writer, dir, base string, controlPlanes, peers []string) error {
 	in := &serverInit{user: "llmgw", stateDir: "/var/lib/llmgw", out: out}
 	if err := in.ensureUser(ctx); err != nil {
 		return err
@@ -105,7 +109,7 @@ func mirrorInit(ctx context.Context, out io.Writer, dir, base string, controlPla
 		}
 		base = "http://" + net.JoinHostPort(addr, "8081")
 	}
-	unit := mirrorUnitFor(dir, ":8082", base, controlPlanes,
+	unit := mirrorUnitFor(dir, ":8082", base, controlPlanes, peers,
 		"'llmgw mirror init'. Bytes on :8081 for every node; the\n# management API on :8082, for the control plane only.")
 	changed, err := in.writeUnit(mirrorUnitPath, unit)
 	if err != nil {

internal/cli/model.go

@@ -18,6 +18,7 @@ import (
 
 	"gitea.mynodes.xyz/ilker/llmgw/internal/control/adminapi"
 	"gitea.mynodes.xyz/ilker/llmgw/internal/control/catalog"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/control/modelserve"
 	"gitea.mynodes.xyz/ilker/llmgw/internal/sourcecheck"
 )
 
@@ -33,7 +34,7 @@ func newModelCmd(client func() *ctlClient, asJSON *bool) *cobra.Command {
 		Use:   "model",
 		Short: "Add or remove models while the cluster is running",
 	}
-	c.AddCommand(newModelAddCmd(client), newModelPresetsCmd(client, asJSON), newModelRemoveCmd(client), newModelPlanCmd(client),
+	c.AddCommand(newModelContextCmd(client), newModelGPUsCmd(client), newModelAddCmd(client), newModelPresetsCmd(client, asJSON), newModelRemoveCmd(client), newModelPlanCmd(client),
 		newResourcesCmd(client), newClassCmd(client), newOwnerCmd(client),
 		newTargetCmd(client, asJSON), newAdaptersCmd(client),
 		newModelServeCmd(client), newModelUpdatesCmd(client, asJSON), newModelUpdateCmd(client),
@@ -386,3 +387,112 @@ func newAdaptersCmd(client func() *ctlClient) *cobra.Command {
 	c.Flags().StringArrayVar(&remove, "remove", nil, "an adapter's name; repeatable")
 	return c
 }
+
+func newModelContextCmd(client func() *ctlClient) *cobra.Command {
+	var slots int
+	c := &cobra.Command{
+		Use:   "context <model> <tokens>",
+		Short: "Change how much one request to a served model may use",
+		Long: "The context lives in three places: the catalog (what a request may use, and\n" +
+			"what it may answer), the engine's arguments (llama.cpp's --ctx-size and\n" +
+			"--parallel, vLLM's --max-model-len), and the memory the KV cache takes. This\n" +
+			"changes all three together. It refuses more than the model was trained for,\n" +
+			"and a context its cards would not hold, saying what each would need. Its\n" +
+			"replicas are then started again on it, each new one before the old stops.\n\n" +
+			"--slots is how many requests a llama.cpp replica answers at once; its pool is\n" +
+			"slots × tokens. Left out, it keeps what it has.",
+		Example: "  llmgw ctl model context qwen3-coder-30b 65536\n" +
+			"  llmgw ctl model context qwen2.5-7b 16384 --slots 4",
+		Args: cobra.ExactArgs(2),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			tokens, err := strconv.Atoi(args[1])
+			if err != nil {
+				return fmt.Errorf("the context is a whole number of tokens, not %q", args[1])
+			}
+			body, _ := json.Marshal(modelserve.ContextRequest{Tokens: tokens, Slots: slots})
+			var v modelserve.ContextChange
+			if err := client().patient(time.Minute).post(cmd.Context(), "/api/v1/models/"+args[0]+"/context", body, &v); err != nil {
+				return err
+			}
+			out := cmd.OutOrStdout()
+			fmt.Fprintf(out, "%s: %d → %d tokens a request, answers up to %d", v.Model, v.Before, v.After, v.MaxOutput)
+			if v.Trained > 0 {
+				fmt.Fprintf(out, " (trained for %d)", v.Trained)
+			}
+			fmt.Fprintln(out)
+			for _, p := range v.Profiles {
+				line := "  " + p.Profile + " (" + p.Engine + "): " + strings.Join(p.Args, " ")
+				switch {
+				case p.Before != "" && p.After != "":
+					line += "; " + p.Before + " → " + p.After
+				case p.After != "":
+					line += "; " + p.After
+				}
+				fmt.Fprintln(out, line)
+			}
+			if v.Note != "" {
+				fmt.Fprintln(out, "  note: "+v.Note)
+			}
+			if v.Restarted > 0 {
+				fmt.Fprintf(out, "%d replica(s) started again on it, each new one before the old stops: llmgw ctl replicas --model %s\n", v.Restarted, v.Model)
+			}
+			return nil
+		},
+	}
+	c.Flags().IntVar(&slots, "slots", 0, "llama.cpp: requests a replica answers at once (default: as it is)")
+	return c
+}
+
+func newModelGPUsCmd(client func() *ctlClient) *cobra.Command {
+	var profile string
+	var inPlace bool
+	c := &cobra.Command{
+		Use:   "gpus <model> <cards>",
+		Short: "Spread a served model over more cards, or fewer",
+		Long: "How many cards of one node each replica uses. vLLM, SGLang and TensorRT-LLM\n" +
+			"spread the model by tensor parallel, which needs the count to divide the\n" +
+			"model's KV heads or be divided by them; llama.cpp splits its layers over the\n" +
+			"cards. It refuses a count a node does not have, and one whose cards would not\n" +
+			"hold the model and its context, saying what each would need. Its replicas are\n" +
+			"then started again on it, each new one beside the old before it stops.\n\n" +
+			"With no node free enough beside the running replicas — a node of two cards,\n" +
+			"the model on one — --in-place stops each replica first: the model does not\n" +
+			"answer while the new one loads.\n\n" +
+			"--profile names the profile for a model with more than one on cards.",
+		Example: "  llmgw ctl model gpus qwen3-coder-30b 2\n" +
+			"  llmgw ctl model gpus llama-70b 4 --profile h100 --in-place",
+		Args: cobra.ExactArgs(2),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			n, err := strconv.Atoi(args[1])
+			if err != nil {
+				return fmt.Errorf("the cards are a whole number, not %q", args[1])
+			}
+			body, _ := json.Marshal(modelserve.GPUsRequest{GPUs: n, Profile: profile, InPlace: inPlace})
+			var v modelserve.GPUsChange
+			if err := client().patient(time.Minute).post(cmd.Context(), "/api/v1/models/"+args[0]+"/gpus", body, &v); err != nil {
+				return err
+			}
+			out := cmd.OutOrStdout()
+			fmt.Fprintf(out, "%s: %d → %d %s card(s) a replica (%s, %s)\n", v.Model, v.Before, v.After, v.Class, v.Profile, v.Engine)
+			if v.PerCard != "" {
+				fmt.Fprintln(out, "  "+v.PerCard)
+			}
+			if len(v.Args) > 0 {
+				fmt.Fprintln(out, "  args: "+strings.Join(v.Args, " "))
+			}
+			if v.Note != "" {
+				fmt.Fprintln(out, "  note: "+v.Note)
+			}
+			switch {
+			case v.Restarted > 0 && v.InPlace:
+				fmt.Fprintf(out, "%d replica(s) stopped and started again on it, one by one: llmgw ctl replicas --model %s\n", v.Restarted, v.Model)
+			case v.Restarted > 0:
+				fmt.Fprintf(out, "%d replica(s) started again on it, each new one before the old stops: llmgw ctl replicas --model %s\n", v.Restarted, v.Model)
+			}
+			return nil
+		},
+	}
+	c.Flags().StringVar(&profile, "profile", "", "the profile to change, for a model with several on cards")
+	c.Flags().BoolVar(&inPlace, "in-place", false, "stop each replica before its replacement starts, when there is no room beside it")
+	return c
+}

internal/cli/node.go

@@ -186,7 +186,7 @@ func (na *nodeAdd) prepare() error {
 	}
 	// Where the mirror serves model files, for the node to be asked to reach.
 	if cfg, err := loadEffectiveConfig(na.serverCfg); err == nil && cfg.Mirror.API != "" {
-		if mc, err := mirror.NewClient(cfg.Mirror.API, cfg.Mirror.TLS, 5*time.Second); err == nil {
+		if mc, err := mirror.NewFailoverClient(append([]string{cfg.Mirror.API}, cfg.Mirror.Peers...), cfg.Mirror.TLS, 5*time.Second); err == nil {
 			ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
 			if list, err := mc.List(ctx); err == nil {
 				o.MirrorBase = list.Base

internal/cli/root.go

@@ -55,6 +55,7 @@ func Root() *cobra.Command {
 		newBundleCmd(),
 		newSimEngineCmd(),
 		newComfyUIGateCmd(),
+		newStrataServeCmd(),
 		newDoctorCmd(),
 		newNodeCmd(),
 		newSelfUpgradeCmd(),

internal/cli/server.go

@@ -567,10 +567,21 @@ func newServerCmd() *cobra.Command {
 			// rather than turning into a broken admin page hours later when
 			// somebody finally opens it.
 			if cfg.Mirror.API != "" {
-				mc, err := mirror.NewClient(cfg.Mirror.API, cfg.Mirror.TLS, cfg.Mirror.Timeout)
+				mirrors := append([]string{cfg.Mirror.API}, cfg.Mirror.Peers...)
+				mc, err := mirror.NewFailoverClient(mirrors, cfg.Mirror.TLS, cfg.Mirror.Timeout)
 				if err != nil {
 					return fmt.Errorf("the model mirror: %w", err)
 				}
+				// Mirrors kept in step: a node is given a model's file on one
+				// that answers while the one the catalog names is down.
+				if len(cfg.Mirror.Peers) > 0 {
+					src, err := mirror.NewSources(mirrors, cfg.Mirror.TLS, log)
+					if err != nil {
+						return fmt.Errorf("the model mirrors: %w", err)
+					}
+					go src.Run(ctx)
+					agentsvc.SetSourceRewrite(src.Rewrite)
+				}
 				rt.SetMirror(mc)
 				if fwPub != nil {
 					fwPub.setMirror(mc)
@@ -578,7 +589,7 @@ func newServerCmd() *cobra.Command {
 				// Adapters trained on the nodes are kept there too; sending one
 				// takes longer than the mirror's usual answers.
 				if adminAPI != nil {
-					if up, err := mirror.NewClient(cfg.Mirror.API, cfg.Mirror.TLS, 10*time.Minute); err == nil {
+					if up, err := mirror.NewFailoverClient(mirrors, cfg.Mirror.TLS, 10*time.Minute); err == nil {
 						adminAPI.SetTrainMirror(up)
 					}
 				}
@@ -596,7 +607,7 @@ func newServerCmd() *cobra.Command {
 						// Updates: the catalog in force, the service that
 						// changes it, and what replaces replicas.
 						Current: func() []catalog.Model { return admin.Catalog().Models() },
-						Models:  catalogSvc, Restart: admin.Reconciler().Restart,
+						Models:  catalogSvc, Restart: admin.Reconciler().Restart, RestartInPlace: admin.Reconciler().RestartInPlace,
 						Pool: adminAPI.PoolNodes, Notify: notifier,
 						// Canaries: how the router saw each side serve.
 						CanaryStats: rt.CanaryStats, ResetCanary: rt.ResetCanary,

internal/cli/serverinit.go

@@ -57,10 +57,12 @@ WantedBy=multi-user.target
 // it again finishes an install that stopped halfway and changes nothing on
 // one that did not.
 type serverInit struct {
-	etcDir   string // /etc/llmgw
-	stateDir string // /var/lib/llmgw
-	pkiDir   string // where the authority lives
-	unitPath string // "" leaves systemd alone
+	// serviceActive says whether llmgw-server runs; nil asks systemctl.
+	serviceActive func() bool
+	etcDir        string // /etc/llmgw
+	stateDir      string // /var/lib/llmgw
+	pkiDir        string // where the authority lives
+	unitPath      string // "" leaves systemd alone
 
 	// user owns the state and reads the key; "" means no user is created and
 	// nothing is chowned, which is what a test wants.
@@ -84,6 +86,9 @@ type serverInit struct {
 	// nodes labelled for them; judgeCPUs their cores, once added.
 	judgeMachines []judgeMachine
 	judgeCPUs     []int
+	// secondMirror is which machine of the store, 1 or 2, keeps a second
+	// mirror in step with this machine's; 0 is none.
+	secondMirror int
 	// judgeNote is what timing the judge found, said before the next
 	// steps when it is not fine.
 	judgeNote string
@@ -241,6 +246,8 @@ func newServerInitCmd() *cobra.Command {
 	f.StringVar(&judgeCats, "judge-categories", strings.Join(defaultJudgeCategories, ","),
 		"what the judge refuses, comma-separated, of: "+strings.Join(judgeCategories, ", "))
 	f.IntVar(&in.judge.Replicas, "judge-replicas", 2, "how many copies of the judge run, each on CPUs")
+	f.IntVar(&in.secondMirror, "second-mirror", 0,
+		"with --patroni, a second mirror on the store's machine 1 or 2, kept in step with this one's (0: none)")
 	f.Var(judgeMachineFlag{&in.judgeMachines}, "judge-node",
 		"user@host of a machine for the guardrails' judges alone: added as a node, and the judges run only there; repeatable")
 	f.DurationVar(&in.judge.Budget, "judge-budget", 5*time.Second, "how long a request waits for the judge")
@@ -355,6 +362,12 @@ func (in *serverInit) run(ctx context.Context) error {
 			return fmt.Errorf("standby control planes: %w", err)
 		}
 	}
+	if hasSystemd && in.secondMirror > 0 {
+		if err := in.ensureSecondMirror(ctx); err != nil {
+			// Not fatal: the first mirror serves; said, and how to try again.
+			fmt.Fprintf(in.out, "  !!    second mirror: %v; later: sudo llmgw mirror add-peer <user>@<machine>\n", err)
+		}
+	}
 	if hasSystemd && in.judgeWanted() && in.configJudgesWith(guardModelName) {
 		in.checkJudge(ctx)
 	}
@@ -365,12 +378,7 @@ func (in *serverInit) run(ctx context.Context) error {
 // judgeWanted is whether this install fetches the judge model, and so
 // whether the baseline policy asks it.
 func (in *serverInit) judgeWanted() bool {
-	if in.noGuardModel || (in.noMirror && in.mirrorURL == "") || in.unitPath == "" {
-		return false
-	}
-	// Offline, only a judge the bundle carries: a policy asking a judge
-	// that never comes refuses every request.
-	if in.bundle != "" && !in.bundleCarries(guardModelName) {
+	if !in.judgeComes() || in.unitPath == "" {
 		return false
 	}
 	_, err := exec.LookPath("systemctl")
@@ -1000,3 +1008,13 @@ func virtualInterface(name string) bool {
 	}
 	return false
 }
+
+// judgeComes is whether this setup has a judge to install: one not left
+// out, a mirror to fetch it from, and, offline, a bundle that carries it —
+// a policy asking a judge that never comes refuses every request.
+func (in *serverInit) judgeComes() bool {
+	if in.noGuardModel || (in.noMirror && in.mirrorURL == "") {
+		return false
+	}
+	return in.bundle == "" || in.bundleCarries(guardModelName)
+}

internal/cli/serverinit_choices.go

@@ -5,6 +5,7 @@ import (
 	"math"
 	"net"
 	"os"
+	"os/exec"
 	"path"
 	"slices"
 	"strconv"
@@ -176,7 +177,9 @@ func (in *serverInit) servingMemoryMB(judgesElsewhere bool) (mb int, what string
 		gb += in.starterModelSpec().minMemoryGB()
 		parts = append(parts, "the starter model")
 	}
-	if !in.noGuardModel && !judgesElsewhere {
+	// Only judges that will be installed: none without a mirror to fetch
+	// them from, or from a bundle that does not carry one.
+	if !judgesElsewhere && in.judgeComes() {
 		n := in.judgeFor().Replicas
 		gb += n * judgeMemoryGB
 		parts = append(parts, fmt.Sprintf("%d judges", n))
@@ -193,6 +196,12 @@ func (in *serverInit) checkMemory() error {
 	if in.unitPath == "" {
 		return nil // a development run, served by nothing
 	}
+	// Run again on a machine already serving — to change --host, say, as
+	// the docs advise — it installs no models it does not already serve. A
+	// first run that stopped half way is checked again: it still would.
+	if in.alreadyServing() {
+		return nil
+	}
 	meminfo, err := os.ReadFile("/proc/meminfo")
 	if err != nil {
 		return nil
@@ -387,6 +396,9 @@ func (in *serverInit) initFlags() []string {
 	for _, m := range in.judgeMachines {
 		add("--judge-node", m.target)
 	}
+	if in.secondMirror > 0 {
+		add("--second-mirror", fmt.Sprint(in.secondMirror))
+	}
 	switch {
 	case in.mirrorURL != "":
 		add("--mirror", in.mirrorURL)
@@ -471,3 +483,15 @@ func shellWord(s string) string {
 	}
 	return "'" + strings.ReplaceAll(s, "'", `'\''`) + "'"
 }
+
+// alreadyServing reports whether this machine's gateway is set up and
+// running: a server.yaml, and its unit active.
+func (in *serverInit) alreadyServing() bool {
+	if _, err := os.Stat(in.configPath()); err != nil {
+		return false
+	}
+	if in.serviceActive != nil {
+		return in.serviceActive()
+	}
+	return exec.Command("systemctl", "is-active", "--quiet", "llmgw-server").Run() == nil
+}

internal/cli/serverinit_cluster.go

@@ -123,10 +123,10 @@ func (in *serverInit) writeUnit(path, body string) (bool, error) {
 func (in *serverInit) mirrorUnit() string {
 	base := "http://" + net.JoinHostPort(in.reachableAddr(), "8081")
 	if in.standbys {
-		return mirrorUnitFor(in.mirrorRoot(), ":8082", base, in.mirrorAllow(),
+		return mirrorUnitFor(in.mirrorRoot(), ":8082", base, in.mirrorAllow(), nil,
 			"'llmgw server init'. Bytes on :8081 for every node; the\n# management API for the control planes only, whichever leads.")
 	}
-	return mirrorUnitFor(in.mirrorRoot(), mirrorAPI, base, nil,
+	return mirrorUnitFor(in.mirrorRoot(), mirrorAPI, base, nil, nil,
 		"'llmgw server init'. Bytes on :8081 for every node; the\n# management API on loopback only, for the control plane beside it.")
 }
 
@@ -154,11 +154,14 @@ func (in *serverInit) mirrorAllow() []string {
 
 // mirrorUnitFor is a mirror's systemd unit. allow is who, besides the
 // machine itself, may use the management API.
-func mirrorUnitFor(dir, apiListen, base string, allow []string, about string) string {
+func mirrorUnitFor(dir, apiListen, base string, allow, peers []string, about string) string {
 	exec := "/usr/local/bin/llmgw mirror serve --dir " + dir + " --listen :8081 --api-listen " + apiListen + " --base " + base
 	for _, a := range allow {
 		exec += " --api-allow " + a
 	}
+	for _, p := range peers {
+		exec += " --peer " + p
+	}
 	return `[Unit]
 Description=llmgw model mirror
 After=network-online.target

internal/cli/serverinit_memory_test.go

@@ -1,6 +1,7 @@
 package cli
 
 import (
+	"os"
 	"testing"
 
 	"gitea.mynodes.xyz/ilker/llmgw/internal/preflight"
@@ -24,4 +25,28 @@ func TestTheMemoryAMachineNeedsCountsWhatItServes(t *testing.T) {
 	if mb, _ := (&serverInit{noModel: true, noGuardModel: true}).servingMemoryMB(false); mb != 0 {
 		t.Errorf("nothing served: %d", mb)
 	}
+	// Without a mirror no judge is fetched, so none is counted.
+	if mb, what := (&serverInit{noMirror: true}).servingMemoryMB(false); mb != 0 {
+		t.Errorf("no mirror, nothing to fetch: %d %q", mb, what)
+	}
+}
+
+func TestSetupRunAgainOnAMachineAlreadySetUpIsNotRefusedItsMemory(t *testing.T) {
+	// The docs fix a certificate with server init --host on a machine set up
+	// before the check: it installs nothing it does not already serve.
+	dir := t.TempDir()
+	cfg := dir + "/server.yaml"
+	if err := os.WriteFile(cfg, []byte("listen: :8080\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	in := &serverInit{unitPath: dir + "/llmgw-server.service", etcDir: dir, serviceActive: func() bool { return true }}
+	if err := in.checkMemory(); err != nil {
+		t.Errorf("refused: %v", err)
+	}
+	// A first run that stopped half way left server.yaml and serves nothing:
+	// it still would install the models, and is checked.
+	in.serviceActive = func() bool { return false }
+	if in.alreadyServing() {
+		t.Error("a machine whose gateway is not running was taken for one serving")
+	}
 }

internal/cli/strataserve.go

@@ -0,0 +1,43 @@
+package cli
+
+import (
+	"os/signal"
+	"syscall"
+
+	"github.com/spf13/cobra"
+
+	"gitea.mynodes.xyz/ilker/llmgw/internal/strata"
+)
+
+func newStrataServeCmd() *cobra.Command {
+	var cfg strata.Config
+	cmd := &cobra.Command{
+		Use:    "strata-serve [flags] [-- setup.py flags]",
+		Short:  "Run Strata as an engine: prepare the model, then serve it (started by the agent)",
+		Hidden: true,
+		Long: "Started by the node agent for a replica whose engine is strata. Prepares the\n" +
+			"model from the fetched GGUF files with Strata's setup, which skips what it has\n" +
+			"done before, then becomes Strata's server on the replica's address. The key is\n" +
+			"read from " + strata.KeyEnv + ".",
+		RunE: func(cmd *cobra.Command, args []string) error {
+			ctx, stop := signal.NotifyContext(cmd.Context(), syscall.SIGINT, syscall.SIGTERM)
+			defer stop()
+			cfg.Args = args
+			cfg.Log = logger()
+			return strata.Run(ctx, cfg)
+		},
+	}
+	f := cmd.Flags()
+	f.StringVar(&cfg.Listen, "listen", "", "the replica's address")
+	f.StringVar(&cfg.Model, "model", "", "the model's name, for /v1/models")
+	f.StringVar(&cfg.Weights, "weights", "", "the fetched GGUF files' directory")
+	f.StringVar(&cfg.State, "state", "", "the replica's own directory")
+	f.StringVar(&cfg.Data, "data", "", "the node's directory for prepared models, shared by its replicas")
+	f.StringVar(&cfg.Python, "python", "python3", "the interpreter of Strata's environment")
+	f.StringVar(&cfg.Dir, "strata", "", "Strata's directory, where setup.py is")
+	f.StringVar(&cfg.Engine, "engine", "", "Strata's engine directory, where the strata binary is")
+	for _, n := range []string{"listen", "weights", "state", "data", "strata", "engine"} {
+		_ = cmd.MarkFlagRequired(n)
+	}
+	return cmd
+}

internal/cli/tui_actions.go

@@ -332,6 +332,55 @@ func (m consoleModel) modelMenu(md adminapi.ModelView) *menuDialog {
 				}})
 			return nil
 		}},
+		{"Context length…", func(m *consoleModel) tea.Cmd {
+			m.openForm(&formDialog{title: "Context of " + md.Name,
+				note: "What one request may use. The engine's arguments and the memory change with it, checked against " +
+					"what the cards hold; its replicas are then started again on it, one by one.",
+				fields: []formField{
+					textF("tokens", strconv.Itoa(md.ContextLength), "a request's prompt and answer together"),
+					textF("slots", "", "llama.cpp: requests at once; empty keeps it"),
+				},
+				submit: func(m *consoleModel, v []string) (tea.Cmd, error) {
+					tokens, err := strconv.Atoi(v[0])
+					if err != nil || tokens < 256 {
+						return nil, fmt.Errorf("the context is a whole number of tokens, at least 256")
+					}
+					slots := 0
+					if v[1] != "" {
+						if slots, err = strconv.Atoi(v[1]); err != nil || slots < 1 {
+							return nil, fmt.Errorf("slots is a whole number from 1")
+						}
+					}
+					return m.call("Context of "+md.Name, http.MethodPost, path+"/context",
+						modelserve.ContextRequest{Tokens: tokens, Slots: slots}, contextSays), nil
+				}})
+			return nil
+		}},
+		{"GPUs…", func(m *consoleModel) tea.Cmd {
+			profiles := append([]string{"(its only one on cards)"}, md.Profiles...)
+			m.openForm(&formDialog{title: "Cards of " + md.Name,
+				note: "How many cards of one node each replica uses: tensor parallel for vLLM, SGLang and TensorRT-LLM, " +
+					"the layers split for llama.cpp. Checked against the model's heads, its cards' memory and the nodes; " +
+					"its replicas are then started again on it, one by one.",
+				fields: []formField{
+					choiceF("profile", profiles, profiles[0]),
+					textF("cards", "", "a replica's, on one node"),
+					choiceF("in place", []string{"no", "yes"}, "no"),
+				},
+				submit: func(m *consoleModel, v []string) (tea.Cmd, error) {
+					n, err := strconv.Atoi(v[1])
+					if err != nil || n < 1 {
+						return nil, fmt.Errorf("the cards are a whole number from 1")
+					}
+					profile := v[0]
+					if profile == profiles[0] {
+						profile = ""
+					}
+					return m.call("Cards of "+md.Name, http.MethodPost, path+"/gpus",
+						modelserve.GPUsRequest{GPUs: n, Profile: profile, InPlace: v[2] == "yes"}, gpusSays), nil
+				}})
+			return nil
+		}},
 		{"Resources…", func(m *consoleModel) tea.Cmd {
 			profiles := md.Profiles
 			if len(profiles) == 0 {
@@ -898,3 +947,46 @@ func locationRequest(n adminapi.NodeView, vals []string) adminapi.NodeLabelsRequ
 	}
 	return req
 }
+
+// contextSays is a context change in a sentence or two.
+func contextSays(raw json.RawMessage) string {
+	var v modelserve.ContextChange
+	if json.Unmarshal(raw, &v) != nil {
+		return "Done."
+	}
+	s := fmt.Sprintf("%s: %d → %d tokens a request, answers up to %d.", v.Model, v.Before, v.After, v.MaxOutput)
+	for _, p := range v.Profiles {
+		if p.After != "" {
+			s += "\n" + p.Profile + ": " + p.After
+		}
+	}
+	if v.Note != "" {
+		s += "\n" + v.Note
+	}
+	if v.Restarted > 0 {
+		s += fmt.Sprintf("\n%d replica(s) started again on it, one by one.", v.Restarted)
+	}
+	return s
+}
+
+// gpusSays is a change of cards in a sentence or two.
+func gpusSays(raw json.RawMessage) string {
+	var v modelserve.GPUsChange
+	if json.Unmarshal(raw, &v) != nil {
+		return "Done."
+	}
+	s := fmt.Sprintf("%s: %d → %d %s card(s) a replica.", v.Model, v.Before, v.After, v.Class)
+	if v.PerCard != "" {
+		s += "\n" + v.PerCard
+	}
+	if v.Note != "" {
+		s += "\n" + v.Note
+	}
+	switch {
+	case v.Restarted > 0 && v.InPlace:
+		s += fmt.Sprintf("\n%d replica(s) stopped and started again on it, one by one.", v.Restarted)
+	case v.Restarted > 0:
+		s += fmt.Sprintf("\n%d replica(s) started again on it, one by one.", v.Restarted)
+	}
+	return s
+}

internal/cli/tui_choices.go

@@ -62,6 +62,7 @@ type installForm struct {
 	admin, hosts, language, signup         *setupField
 	storeKind                              *setupField
 	mirrorWhere, mirrorDirF, mirrorURL     *setupField
+	secondMirror                           *setupField
 	starterPick, starterSource, starterCtx *setupField
 	judgeFail, judgeCopies, judgeBudget    *setupField
 	judgeCats                              []*setupField
@@ -181,18 +182,26 @@ func newInstallForm(in *serverInit, host string, found []string, ha bool) *insta
 	f.mirrorDirF.Show = func() bool { return f.mirrorWhere.Choice == 0 }
 	f.mirrorURL = textSetupField("Its management API", in.mirrorURL, "http://10.0.0.20:8082", "")
 	f.mirrorURL.Show = func() bool { return f.mirrorWhere.Choice == 1 }
+	f.secondMirror = choiceField("A second mirror", []string{"none", "on database machine 1", "on database machine 2"},
+		in.secondMirror, "Kept in step with this one, each copying what the other has: while either machine is down, "+
+			"models are still added and fetched. It needs room for every model, as this one does.")
+	f.secondMirror.Show = func() bool { return ha && f.mirrorWhere.Choice == 0 }
 	f.mirror = &installPart{
 		title: "Model mirror", on: !in.noMirror || in.mirrorURL != "",
 		about: "Downloads each model once, from Hugging Face or wherever it comes from, and serves the files to " +
 			"every node on the network, so ten nodes do not download the same 40 GB ten times. " +
 			"The starter model, the judge and an offline bundle go into it.",
-		fields: []*setupField{f.mirrorWhere, f.mirrorDirF, f.mirrorURL},
+		fields: []*setupField{f.mirrorWhere, f.mirrorDirF, f.mirrorURL, f.secondMirror},
 		off:    "each model's files must already be on a mirror somewhere; no starter model, no judge",
 		summary: func() string {
 			if f.mirrorWhere.Choice == 1 {
 				return "at " + orText(f.mirrorURL.Value(), "(an address)")
 			}
-			return "on this machine, in " + orText(f.mirrorDirF.Value(), mirrorDir)
+			s := "on this machine, in " + orText(f.mirrorDirF.Value(), mirrorDir)
+			if ha && f.secondMirror.Choice > 0 {
+				s += "; a second " + f.secondMirror.Options[f.secondMirror.Choice]
+			}
+			return s
 		},
 	}
 
@@ -636,6 +645,10 @@ func (f *installForm) apply(in *serverInit) error {
 	in.postgres = !f.ha && f.storeKind.Choice == 1
 	in.noMirror = !f.mirror.on || mirrorURL != ""
 	in.mirrorURL, in.mirrorPath = mirrorURL, mirrorPath
+	in.secondMirror = 0
+	if f.ha && f.mirror.on && f.mirrorWhere.Choice == 0 {
+		in.secondMirror = f.secondMirror.Choice
+	}
 	in.noNode = !f.node.on
 	in.noModel = !f.starter.on
 	in.starter, in.starterCtx = starter, ctx

internal/cli/tui_console.go

@@ -46,6 +46,7 @@ const (
 	cQuotas
 	cGuard
 	cKB
+	cMCP
 	cFederation
 	cMirror
 	cJobs
@@ -70,7 +71,7 @@ const (
 	cCount
 )
 
-var consoleLabels = [cCount]string{"Overview", "Monitor", "Nodes", "GPUs", "Models", "Replicas", "Usage", "Quotas", "Guardrails", "Knowledge", "Federation", "Mirror", "Jobs", "Logs", "Accounts", "API keys", "Organizations", "Sign-in", "Audit", "Manual",
+var consoleLabels = [cCount]string{"Overview", "Monitor", "Nodes", "GPUs", "Models", "Replicas", "Usage", "Quotas", "Guardrails", "Knowledge", "MCP servers", "Federation", "Mirror", "Jobs", "Logs", "Accounts", "API keys", "Organizations", "Sign-in", "Audit", "Manual",
 	"server.yaml", "Add a node", "Add a control plane", "Replicate the store", "Serve a model", "Serve the guardrails' judge", "Upgrade the cluster",
 	"Restart all replicas", "Back up now", "Wipe this machine", "Leave"}
 
@@ -184,6 +185,7 @@ type consoleModel struct {
 	guard     guardState
 	signIn    signInState
 	kb        kbState
+	mcp       mcpState
 
 	st     *storeHolder
 	people peopleData
@@ -342,6 +344,9 @@ func (m consoleModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 		if m.menu == cKB {
 			cmds = append(cmds, m.fetchKB())
 		}
+		if m.menu == cMCP {
+			cmds = append(cmds, m.fetchMCP())
+		}
 		if m.menu == cMirror {
 			cmds = append(cmds, m.fetchMirror())
 		}
@@ -404,6 +409,8 @@ func (m consoleModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 		m.quotas, m.quotasErr = msg.rows, msg.err
 	case consoleKBMsg:
 		m.kb = kbState{view: msg.view, err: msg.err, loaded: true}
+	case consoleMCPMsg:
+		m.mcp = mcpState(msg)
 	case consoleSignInMsg:
 		m.signIn = signInState{view: msg.view, err: msg.err, loaded: true, ldap: msg.ldap, oidc: msg.oidc}
 	case consoleGuardMsg:
@@ -568,6 +575,8 @@ func (m consoleModel) key(k tea.KeyMsg) (tea.Model, tea.Cmd) {
 			m.menuD = m.signInMenu(m.signInRowKey())
 		case cKB:
 			m.menuD = m.kbMenu(m.kb.view.Collections[m.row])
+		case cMCP:
+			m.menuD = m.mcpMenu(m.mcp.servers[m.row])
 		case cMirror:
 			m.menuD = m.mirrorMenu(m.mirrorGroups()[m.row])
 		case cJobs:
@@ -633,6 +642,10 @@ func (m consoleModel) key(k tea.KeyMsg) (tea.Model, tea.Cmd) {
 			if m.kb.view.Available {
 				m.kbCreateForm()
 			}
+		case cMCP:
+			if m.mcp.loaded && m.mcp.err == nil {
+				m.mcpAddForm()
+			}
 		case cModels:
 			m.addModelForm()
 		case cMirror:
@@ -705,6 +718,8 @@ func (m consoleModel) onEnter() tea.Cmd {
 		return m.fetchSignIn()
 	case cKB:
 		return m.fetchKB()
+	case cMCP:
+		return m.fetchMCP()
 	case cMirror:
 		return m.fetchMirror()
 	case cMonitor:
@@ -1085,6 +1100,8 @@ func (m consoleModel) rowCount() int {
 		return len(signInRows)
 	case cKB:
 		return len(m.kb.view.Collections)
+	case cMCP:
+		return len(m.mcp.servers)
 	case cMirror:
 		return len(m.mirrorGroups())
 	case cJobs:
@@ -1326,6 +1343,8 @@ func (m consoleModel) panel() (title, body, keys string) {
 		return m.signInPanel(pw)
 	case cKB:
 		return m.kbPanel(pw)
+	case cMCP:
+		return m.mcpPanel(pw)
 	case cAccounts, cKeys, cOrgs:
 		return m.peoplePanel(pw)
 	case cManual:

internal/cli/tui_mcp.go

@@ -0,0 +1,534 @@
+package cli
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/url"
+	"slices"
+	"strconv"
+	"strings"
+	"time"
+
+	tea "github.com/charmbracelet/bubbletea"
+
+	"gitea.mynodes.xyz/ilker/llmgw/internal/control/adminapi"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/store"
+)
+
+// The company's own MCP servers, as the console shows them: each with its
+// address, the names of the credentials it is sent (never their values),
+// its rate, and who has been granted its tools; Enter for what can be done
+// to one, A to add one. Clients reach them all through the gateway's /mcp
+// with an llmgw key, as server.tool.
+
+type mcpState struct {
+	servers []adminapi.MCPServerView
+	err     error
+	loaded  bool
+	// grants are the tool grants in the tenancy tree, nil when the store
+	// could not be read from here; where says what each scope's ID is
+	// called, and ids the other way, for the grant form.
+	grants []mcpGrant
+	where  map[string]string
+	ids    map[string]store.LimitOwner
+	// keys are the IDs of the keys there are, for a grant to one.
+	keys map[string]bool
+}
+
+// mcpGrant is one level of the tree that grants tools, and which.
+type mcpGrant struct {
+	owner store.LimitOwner
+	label string
+	tools []string
+}
+
+type consoleMCPMsg mcpState
+
+func (m consoleModel) fetchMCP() tea.Cmd {
+	c, h := m.client, m.st
+	return func() tea.Msg {
+		ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+		defer cancel()
+		var msg consoleMCPMsg
+		msg.err = c.get(ctx, "/api/v1/mcp/servers", &msg.servers)
+		msg.loaded = true
+		if h == nil {
+			return msg
+		}
+		s, err := h.get(ctx)
+		if err != nil {
+			return msg
+		}
+		limits, err := s.AllLimits(ctx)
+		if err != nil {
+			return msg
+		}
+		msg.where, msg.ids = map[string]string{}, map[string]store.LimitOwner{}
+		if tree, err := s.Tree(ctx); err == nil {
+			for _, o := range tree {
+				msg.name(store.ScopeOrg, o.ID, o.Name)
+				for _, t := range o.Teams {
+					msg.name(store.ScopeTeam, t.ID, o.Name+"/"+t.Name)
+					for _, p := range t.Projects {
+						msg.name(store.ScopeProject, p.ID, o.Name+"/"+t.Name+"/"+p.Name)
+					}
+				}
+			}
+		}
+		if users, err := s.Users(ctx, ""); err == nil {
+			for _, u := range users {
+				msg.name(store.ScopeUser, u.ID, u.Email)
+			}
+		}
+		msg.keys = map[string]bool{}
+		if keys, err := s.AllKeys(ctx); err == nil {
+			for _, k := range keys {
+				msg.keys[k.ID] = true
+				msg.where[store.LimitOwner{Scope: store.ScopeKey, ID: k.ID}.CounterKey()] = k.Name
+			}
+		}
+		msg.grants = []mcpGrant{}
+		for _, l := range limits {
+			if l.AllowedTools == nil {
+				continue
+			}
+			o := store.LimitOwner{Scope: l.Scope, ID: l.ScopeID}
+			label := msg.where[o.CounterKey()]
+			if label == "" {
+				label = l.ScopeID
+			}
+			msg.grants = append(msg.grants, mcpGrant{owner: o, label: string(l.Scope) + " " + label, tools: l.AllowedTools})
+		}
+		slices.SortFunc(msg.grants, func(a, b mcpGrant) int { return strings.Compare(a.label, b.label) })
+		return msg
+	}
+}
+
+func (s *consoleMCPMsg) name(scope store.Scope, id, label string) {
+	o := store.LimitOwner{Scope: scope, ID: id}
+	s.where[o.CounterKey()] = label
+	s.ids[string(scope)+":"+strings.ToLower(label)] = o
+	if m := store.MemberScope(scope); m != "" {
+		s.where[store.LimitOwner{Scope: m, ID: id}.CounterKey()] = label
+	}
+}
+
+// closed are the levels whose list names no tool at all.
+func (k mcpState) closed() []mcpGrant {
+	var out []mcpGrant
+	for _, g := range k.grants {
+		if len(g.tools) == 0 {
+			out = append(out, g)
+		}
+	}
+	return out
+}
+
+// mcpGrantsOf is who may use a server's tools, and which of them.
+func (k mcpState) grantsOf(server string) []mcpGrant {
+	var out []mcpGrant
+	for _, g := range k.grants {
+		var mine []string
+		for _, t := range g.tools {
+			if strings.HasPrefix(t, server+".") || t == "*" {
+				mine = append(mine, t)
+			}
+		}
+		if len(mine) > 0 {
+			out = append(out, mcpGrant{owner: g.owner, label: g.label, tools: mine})
+		}
+	}
+	return out
+}
+
+func (m consoleModel) mcpPanel(pw int) (string, string, string) {
+	keys := "→ or Enter open · ↑↓ choose · Q leave"
+	if m.inPanel {
+		keys = "↑↓ server · Enter actions · A add a server · ← back · Q leave"
+	}
+	k := m.mcp
+	var b strings.Builder
+	b.WriteString("\n  " + sDim.Render("The company's MCP servers, offered to agents through the gateway's /mcp as server.tool.") + "\n\n")
+	switch {
+	case k.err != nil:
+		b.WriteString("  " + sBad.Render(k.err.Error()))
+		return "MCP servers", b.String(), keys
+	case !k.loaded:
+		b.WriteString("  " + sDim.Render("asking the control plane…"))
+		return "MCP servers", b.String(), keys
+	case len(k.servers) == 0:
+		b.WriteString("  " + sDim.Render("None yet. A to add one: its address, and the credential it wants."))
+		b.WriteString("\n\n" + mcpClientHint())
+		return "MCP servers", b.String(), keys
+	}
+	var rows [][]string
+	for _, s := range k.servers {
+		state := sOK.Render("on")
+		if !s.Enabled {
+			state = sDim.Render("off")
+		}
+		granted := "—"
+		if k.grants != nil {
+			var who []string
+			for _, g := range k.grantsOf(s.Name) {
+				who = append(who, g.label)
+			}
+			granted = sBad.Render("nobody yet")
+			if len(who) > 0 {
+				granted = strings.Join(who, ", ")
+			}
+		}
+		rate := "—"
+		if s.RatePerMinute > 0 {
+			rate = fmt.Sprintf("%d/min", s.RatePerMinute)
+		}
+		rows = append(rows, []string{s.Name, state, granted, rate, dash(strings.Join(mcpCredentials(s), ", ")), s.URL})
+	}
+	b.WriteString(table([]string{"SERVER", "STATE", "GRANTED TO", "RATE", "SENT", "ADDRESS"}, rows, m.row, pw, m.inPanel))
+	b.WriteString("\n\n  " + sDim.Render("A server's tools are seen only where they are granted; nothing is granted by default.") + "\n")
+	if closed := k.closed(); len(closed) > 0 {
+		var names []string
+		for _, g := range closed {
+			names = append(names, g.label)
+		}
+		b.WriteString("  " + sBad.Render("Closed to every tool: ") + strings.Join(names, ", ") +
+			sDim.Render(" (a server's menu opens them)") + "\n")
+	}
+	b.WriteString("\n")
+	b.WriteString(mcpClientHint())
+	return "MCP servers", b.String(), keys
+}
+
+// mcpClientHint is how a person's agent is pointed at the gateway.
+func mcpClientHint() string {
+	return "  " + sDim.Render("In a person's agent: the gateway's https://…/mcp, with their key from /profile/keys") + "\n" +
+		"  " + sDim.Render("  claude mcp add --transport http company https://HOST/mcp --header \"Authorization: Bearer sk-llmgw-…\"")
+}
+
+// mcpCredentials are the names of the headers a server is sent: which are
+// kept as secrets and which as written. Never their values.
+func mcpCredentials(s adminapi.MCPServerView) []string {
+	var out []string
+	for h := range s.HeaderRefs {
+		out = append(out, h+" (secret)")
+	}
+	for h := range s.Headers {
+		out = append(out, h)
+	}
+	slices.Sort(out)
+	return out
+}
+
+// mcpPath is a server's admin API address.
+func mcpPath(name string) string { return "/api/v1/mcp/servers/" + url.PathEscape(name) }
+
+// mcpAuth turns the form's credential fields into what the server is sent:
+// a bearer token, another header, or nothing.
+func mcpAuth(kind, header, secret string) (map[string]string, error) {
+	switch kind {
+	case "none":
+		return nil, nil
+	case "bearer":
+		if secret == "" {
+			return nil, nil
+		}
+		return map[string]string{"Authorization": "Bearer " + strings.TrimPrefix(secret, "Bearer ")}, nil
+	default:
+		if header == "" {
+			return nil, fmt.Errorf("which header carries the credential?")
+		}
+		if secret == "" {
+			return nil, nil
+		}
+		return map[string]string{header: secret}, nil
+	}
+}
+
+func atoiOr(s string, def int) (int, error) {
+	if strings.TrimSpace(s) == "" {
+		return def, nil
+	}
+	return strconv.Atoi(strings.TrimSpace(s))
+}
+
+// mcpAddForm registers a server.
+func (m *consoleModel) mcpAddForm() {
+	m.openForm(&formDialog{title: "Add an MCP server",
+		note: "Reached by the control plane over HTTP (streamable HTTP or SSE). Its credential is kept in\n" +
+			"the store, sealed, and never reaches a client. Its tools are seen nowhere until granted.",
+		fields: []formField{
+			textF("name", "", "lowercase, the tools' prefix: jira → jira.search"),
+			textF("address", "", "https://jira-mcp.internal/mcp"),
+			textF("about", "", "what it is for, in a line"),
+			choiceF("credential", []string{"bearer", "header", "none"}, "bearer"),
+			textF("header", "", "for credential: header, its name (X-Api-Key)"),
+			pwF("secret", "the token or key; kept sealed in the store"),
+			textF("timeout", "", "seconds a call may take; empty is the default"),
+			textF("rate", "", "calls a minute per caller; empty is no bound"),
+		},
+		submit: func(m *consoleModel, v []string) (tea.Cmd, error) {
+			name, addr := strings.TrimSpace(v[0]), strings.TrimSpace(v[1])
+			if name == "" || addr == "" {
+				return nil, fmt.Errorf("a name and an address")
+			}
+			if err := adminapi.CheckMCPAddress(addr); err != nil {
+				return nil, err
+			}
+			secrets, err := mcpAuth(v[3], strings.TrimSpace(v[4]), v[5])
+			if err != nil {
+				return nil, err
+			}
+			timeout, err := atoiOr(v[6], 0)
+			if err != nil || timeout < 0 {
+				return nil, fmt.Errorf("timeout is whole seconds")
+			}
+			rate, err := atoiOr(v[7], 0)
+			if err != nil || rate < 0 {
+				return nil, fmt.Errorf("rate is calls a minute")
+			}
+			on := true
+			req := adminapi.MCPServerRequest{URL: addr, Description: v[2], Secrets: secrets, Enabled: &on,
+				TimeoutSeconds: timeout, RatePerMinute: rate}
+			return m.call("Add "+name, http.MethodPut, mcpPath(name), req, func(json.RawMessage) string {
+				return "Added " + name + ". Test it from its menu, then grant its tools: nobody sees them yet."
+			}), nil
+		}})
+}
+
+// mcpMenu is what can be done to one server.
+func (m consoleModel) mcpMenu(s adminapi.MCPServerView) *menuDialog {
+	items := []menuItem{
+		{"Test, and list its tools", func(m *consoleModel) tea.Cmd { return m.mcpTest(s) }},
+		{"Grant its tools to…", func(m *consoleModel) tea.Cmd { m.mcpGrantForm(s); return nil }},
+	}
+	for _, g := range m.mcp.grantsOf(s.Name) {
+		if !slices.ContainsFunc(g.tools, func(t string) bool { return t != "*" }) {
+			continue // every tool, granted with the Limits form; revoked there too
+		}
+		items = append(items, menuItem{"Revoke from " + g.label + " (" + strings.Join(g.tools, ", ") + ")…",
+			func(m *consoleModel) tea.Cmd {
+				m.confirmD = &confirmDialog{title: "Revoke " + s.Name + " from " + g.label,
+					text: "Taken from it and from everything beneath it. Its other grants stay; left naming\n" +
+						"no tool, it is closed rather than opened to what the levels above allow.",
+					yes: func(m *consoleModel) tea.Cmd { return m.mcpSetTools(s.Name, g.owner, nil) }}
+				return nil
+			}})
+	}
+	for _, g := range m.mcp.closed() {
+		items = append(items, menuItem{"Open " + g.label + ", closed to every tool…", func(m *consoleModel) tea.Cmd {
+			m.confirmD = &confirmDialog{title: "Open " + g.label,
+				text: "It then names no tools of its own and sees what the levels above it allow.",
+				yes:  func(m *consoleModel) tea.Cmd { return m.mcpOpen(g.owner) }}
+			return nil
+		}})
+	}
+	items = append(items,
+		menuItem{"Change…", func(m *consoleModel) tea.Cmd { m.mcpEditForm(s); return nil }},
+		menuItem{"Replace its credential…", func(m *consoleModel) tea.Cmd { m.mcpCredentialForm(s); return nil }},
+	)
+	if s.Enabled {
+		items = append(items, menuItem{"Turn off", func(m *consoleModel) tea.Cmd {
+			off := false
+			return m.call("Turn off "+s.Name, http.MethodPut, mcpPath(s.Name), adminapi.MCPServerRequest{URL: s.URL, Enabled: &off},
+				func(json.RawMessage) string {
+					return s.Name + " is off: its tools are listed nowhere and every call is refused."
+				})
+		}})
+	} else {
+		items = append(items, menuItem{"Turn on", func(m *consoleModel) tea.Cmd {
+			on := true
+			return m.call("Turn on "+s.Name, http.MethodPut, mcpPath(s.Name), adminapi.MCPServerRequest{URL: s.URL, Enabled: &on}, nil)
+		}})
+	}
+	items = append(items, menuItem{"Remove…", func(m *consoleModel) tea.Cmd {
+		m.confirmD = &confirmDialog{title: "Remove " + s.Name,
+			text: "Its tools go from every agent at once, and its sealed credential with it.\nThe grants stay in the tree until revoked.",
+			yes: func(m *consoleModel) tea.Cmd {
+				return m.call("Remove "+s.Name, http.MethodDelete, mcpPath(s.Name), nil, nil)
+			}}
+		return nil
+	}})
+	return &menuDialog{title: s.Name + " · " + s.URL, items: items}
+}
+
+// mcpTest asks the server for its tools, with its credentials, as a call
+// through the gateway would.
+func (m *consoleModel) mcpTest(s adminapi.MCPServerView) tea.Cmd {
+	return m.call("Test "+s.Name, http.MethodPost, mcpPath(s.Name)+"/test", nil, func(raw json.RawMessage) string {
+		var t adminapi.MCPTestView
+		_ = json.Unmarshal(raw, &t)
+		if !t.OK {
+			return "It did not answer: " + t.Error
+		}
+		if len(t.Tools) == 0 {
+			return "It answers, and offers no tools."
+		}
+		return fmt.Sprintf("It answers, with %d tools:\n\n  %s", len(t.Tools), strings.Join(t.Tools, "\n  "))
+	})
+}
+
+func (m *consoleModel) mcpEditForm(s adminapi.MCPServerView) {
+	rate := ""
+	if s.RatePerMinute > 0 {
+		rate = strconv.Itoa(s.RatePerMinute)
+	}
+	timeout := ""
+	if s.TimeoutSeconds > 0 {
+		timeout = strconv.Itoa(s.TimeoutSeconds)
+	}
+	m.openForm(&formDialog{title: "Change " + s.Name,
+		note: "Its credential stays as it is; Replace its credential for that.",
+		fields: []formField{
+			textF("address", s.URL, ""),
+			textF("about", s.Description, ""),
+			textF("timeout", timeout, "seconds a call may take"),
+			textF("rate", rate, "calls a minute per caller; 0 is no bound"),
+		},
+		submit: func(m *consoleModel, v []string) (tea.Cmd, error) {
+			if err := adminapi.CheckMCPAddress(strings.TrimSpace(v[0])); err != nil {
+				return nil, err
+			}
+			timeout, err := atoiOr(v[2], 0)
+			if err != nil || timeout < 0 {
+				return nil, fmt.Errorf("timeout is whole seconds")
+			}
+			rate, err := atoiOr(v[3], 0)
+			if err != nil || rate < 0 {
+				return nil, fmt.Errorf("rate is calls a minute")
+			}
+			req := adminapi.MCPServerRequest{URL: strings.TrimSpace(v[0]), Description: v[1], TimeoutSeconds: timeout,
+				RatePerMinute: rate}
+			// An emptied field empties it, rather than leaving it as it was.
+			if strings.TrimSpace(v[1]) == "" {
+				req.Clear = append(req.Clear, "description")
+			}
+			if timeout == 0 {
+				req.Clear = append(req.Clear, "timeout")
+			}
+			if rate == 0 {
+				req.Clear = append(req.Clear, "rate")
+			}
+			return m.call("Change "+s.Name, http.MethodPut, mcpPath(s.Name), req, nil), nil
+		}})
+}
+
+func (m *consoleModel) mcpCredentialForm(s adminapi.MCPServerView) {
+	m.openForm(&formDialog{title: "Replace the credential " + s.Name + " is sent",
+		note: "Kept in the store, sealed. The old value is overwritten, not shown.",
+		fields: []formField{
+			choiceF("credential", []string{"bearer", "header"}, "bearer"),
+			textF("header", "", "for credential: header, its name (X-Api-Key)"),
+			pwF("secret", "the token or key"),
+		},
+		submit: func(m *consoleModel, v []string) (tea.Cmd, error) {
+			if v[2] == "" {
+				return nil, fmt.Errorf("the new secret?")
+			}
+			secrets, err := mcpAuth(v[0], strings.TrimSpace(v[1]), v[2])
+			if err != nil {
+				return nil, err
+			}
+			return m.call("Credential of "+s.Name, http.MethodPut, mcpPath(s.Name),
+				adminapi.MCPServerRequest{URL: s.URL, Secrets: secrets, HeaderRefs: map[string]string{}}, nil), nil
+		}})
+}
+
+// mcpGrantForm grants a server's tools to part of the tree, a person or a
+// key, beside whatever else that level is granted.
+func (m *consoleModel) mcpGrantForm(s adminapi.MCPServerView) {
+	if m.mcp.ids == nil {
+		m.dialog, m.dialogTitle = "error", "Grant "+s.Name
+		m.dialogText = "The store cannot be read from this machine, so its grants cannot be changed here.\n" +
+			"On the control plane: llmgw keys limits --scope team --id ORG/TEAM --tools '" + s.Name + ".*'"
+		return
+	}
+	m.openForm(&formDialog{title: "Grant " + s.Name + "'s tools",
+		note: "Seen there, and below, where every level above that names tools names these too.\n" +
+			"A level given its first grant keeps what it saw from above; other grants stay.",
+		fields: []formField{
+			choiceF("to", []string{"team", "project", "org", "user", "key"}, "team"),
+			textF("which", "", "ORG/TEAM/PROJECT, ORG/TEAM, ORG, an e-mail, or a key's ID"),
+			textF("tools", s.Name+".*", "patterns: "+s.Name+".* for all, or "+s.Name+".search,"+s.Name+".get"),
+		},
+		submit: func(m *consoleModel, v []string) (tea.Cmd, error) {
+			which := strings.TrimSpace(v[1])
+			if which == "" {
+				return nil, fmt.Errorf("which %s?", v[0])
+			}
+			owner, ok := m.mcp.ids[v[0]+":"+strings.ToLower(which)]
+			if v[0] == "key" {
+				owner, ok = store.LimitOwner{Scope: store.ScopeKey, ID: which}, m.mcp.keys[which]
+			}
+			if !ok {
+				return nil, fmt.Errorf("no %s %q", v[0], which)
+			}
+			var tools []string
+			for _, t := range strings.Split(v[2], ",") {
+				if t = strings.TrimSpace(t); t != "" {
+					if !strings.HasPrefix(t, s.Name+".") {
+						return nil, fmt.Errorf("%q is not one of %s's tools: they start %s.", t, s.Name, s.Name)
+					}
+					tools = append(tools, t)
+				}
+			}
+			if len(tools) == 0 {
+				return nil, fmt.Errorf("which tools? %s.* for all", s.Name)
+			}
+			return m.mcpSetTools(s.Name, owner, tools), nil
+		}})
+}
+
+// mcpSetTools grants a level these of a server's tools, or with none takes
+// them away, in the store (store.GrantTools, store.RevokeTools): read and
+// written at once, so another administrator's change since this screen
+// was loaded is kept.
+func (m *consoleModel) mcpSetTools(server string, owner store.LimitOwner, tools []string) tea.Cmd {
+	h, where := m.st, m.mcp.where
+	title := "Grant " + server
+	if len(tools) == 0 {
+		title = "Revoke " + server
+	}
+	return func() tea.Msg {
+		ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+		defer cancel()
+		s, err := h.get(ctx)
+		if err != nil {
+			return actionDoneMsg{title: title, err: err}
+		}
+		if len(tools) == 0 {
+			if err := s.RevokeTools(ctx, owner.Scope, owner.ID, server); err != nil {
+				return actionDoneMsg{title: title, err: err}
+			}
+			return actionDoneMsg{title: title, text: server + "'s tools taken back there; its other grants stay."}
+		}
+		g, err := s.GrantTools(ctx, owner.Scope, owner.ID, server, tools)
+		if err != nil {
+			return actionDoneMsg{title: title, err: err}
+		}
+		text := "Granted " + strings.Join(tools, ", ") + "."
+		for _, b := range g.BlockedBy {
+			text += "\n\nBut " + string(b.Scope) + " " + or(where[b.CounterKey()], b.ID) + " names tools and not these:\n" +
+				server + "'s stay out of reach there until it grants them too."
+		}
+		return actionDoneMsg{title: title, text: text}
+	}
+}
+
+// mcpOpen has a closed level name no tools of its own (store.OpenTools).
+func (m *consoleModel) mcpOpen(owner store.LimitOwner) tea.Cmd {
+	h := m.st
+	return func() tea.Msg {
+		ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+		defer cancel()
+		s, err := h.get(ctx)
+		if err == nil {
+			err = s.OpenTools(ctx, owner.Scope, owner.ID)
+		}
+		if err != nil {
+			return actionDoneMsg{title: "Open", err: err}
+		}
+		return actionDoneMsg{title: "Open", text: "Opened: it names no tools of its own now."}
+	}
+}

internal/cli/tui_mcp_test.go

@@ -0,0 +1,134 @@
+package cli
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"slices"
+	"strings"
+	"testing"
+
+	tea "github.com/charmbracelet/bubbletea"
+
+	"gitea.mynodes.xyz/ilker/llmgw/internal/config"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/control/adminapi"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/store"
+)
+
+func TestAnMCPServerIsAddedAndTestedFromTheConsole(t *testing.T) {
+	m, s := consoleWith(t, cMCP)
+	m.mcp = mcpState{loaded: true, grants: []mcpGrant{}, servers: []adminapi.MCPServerView{{
+		Name: "jira", URL: "https://jira-mcp.internal/mcp", Enabled: true, RatePerMinute: 30,
+		HeaderRefs: map[string]string{"Authorization": "store:mcp/jira/Authorization"},
+	}}}
+	var tm tea.Model = m
+	view := tm.View()
+	for _, want := range []string{"MCP servers", "jira", "Authorization (secret)",
+		"30/min", "nobody yet", "claude mcp add"} {
+		if !strings.Contains(view, want) {
+			t.Errorf("missing %q in:\n%s", want, view)
+		}
+	}
+	if strings.Contains(view, "store:mcp/jira") {
+		t.Error("where the secret is kept is shown")
+	}
+
+	// A: a server with a bearer token, which is sent once and kept sealed.
+	tm = press(tm, "a")
+	f := tm.(consoleModel).formD
+	if f == nil {
+		t.Fatal("no form")
+	}
+	f.fields[0].text.SetValue("hr")
+	f.fields[1].text.SetValue("https://hr-mcp.internal/mcp")
+	f.fields[5].text.SetValue("t0ken")
+	f.focus = len(f.fields)
+	_ = pressRun(tm, "enter")
+	got := s.last()
+	var req adminapi.MCPServerRequest
+	if !strings.HasPrefix(got, "PUT /api/v1/mcp/servers/hr ") ||
+		json.Unmarshal([]byte(strings.SplitN(got, " ", 3)[2]), &req) != nil ||
+		req.URL != "https://hr-mcp.internal/mcp" || req.Secrets["Authorization"] != "Bearer t0ken" ||
+		req.Enabled == nil || !*req.Enabled {
+		t.Fatalf("sent %q", got)
+	}
+
+	// Enter on jira: test it, then turn it off.
+	tm = press(m, "enter")
+	md := tm.(consoleModel).menuD
+	if md == nil || md.items[0].label != "Test, and list its tools" || !strings.Contains(md.title, "https://jira-mcp.internal/mcp") {
+		t.Fatalf("menu %+v", md)
+	}
+	_ = pressRun(tm, "enter")
+	if got := s.last(); got != "POST /api/v1/mcp/servers/jira/test " {
+		t.Fatalf("sent %q", got)
+	}
+}
+
+func TestAnMCPServersToolsAreGrantedBesideOtherGrants(t *testing.T) {
+	// The servers, from a control plane; the grants, from the store.
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_ = json.NewEncoder(w).Encode([]adminapi.MCPServerView{
+			{Name: "hr", URL: "https://hr/mcp", Enabled: true}, {Name: "jira", URL: "https://jira/mcp", Enabled: true}})
+	}))
+	defer srv.Close()
+	dir := t.TempDir()
+	cfgPath := dir + "/server.yaml"
+	if err := os.WriteFile(cfgPath, []byte("auth:\n  store: "+dir+"/llmgw.db\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := config.Load(cfgPath); err != nil {
+		t.Fatal(err)
+	}
+	m := newConsoleModel(srv.URL, cfgPath)
+	defer m.st.close()
+	m.loaded, m.inPanel, m.menu = true, true, cMCP
+	ctx := context.Background()
+	st, err := m.st.get(ctx)
+	if err != nil {
+		t.Fatal(err)
+	}
+	org, _ := st.CreateOrg(ctx, "Acme")
+	team, _ := st.CreateTeam(ctx, org, "ml")
+	if err := st.PutLimits(ctx, store.Limits{Scope: store.ScopeTeam, ScopeID: team, AllowedTools: []string{"jira.*"}}); err != nil {
+		t.Fatal(err)
+	}
+	toolsOf := func() []string {
+		all, _ := st.AllLimits(ctx)
+		return all[store.LimitOwner{Scope: store.ScopeTeam, ID: team}.CounterKey()].AllowedTools
+	}
+	refresh := func(tm tea.Model) tea.Model {
+		tm, _ = tm.Update(tm.(consoleModel).fetchMCP()())
+		return tm
+	}
+
+	var tm tea.Model = refresh(m)
+	if v := tm.View(); !strings.Contains(v, "team Acme/ml") || !strings.Contains(v, "nobody yet") {
+		t.Fatalf("jira's grant, and hr's none, should show:\n%s", v)
+	}
+	// hr is the first row: grant its tools to the team.
+	tm = press(tm, "enter", "down", "enter")
+	f := tm.(consoleModel).formD
+	if f == nil || f.fields[2].value() != "hr.*" {
+		t.Fatalf("no grant form, or not for hr: %+v", f)
+	}
+	f.fields[1].text.SetValue("Acme/ml")
+	f.focus = len(f.fields)
+	tm = pressRun(tm, "enter")
+	if got := toolsOf(); !slices.Equal(got, []string{"jira.*", "hr.*"}) {
+		t.Fatalf("the team's grants are %v; jira's should have stayed beside hr's", got)
+	}
+
+	// Revoke hr from the team: jira's stays.
+	tm = refresh(press(tm, "x"))
+	tm = press(tm, "enter", "down", "down", "enter")
+	if c := tm.(consoleModel).confirmD; c == nil || !strings.Contains(c.title, "Revoke hr from team Acme/ml") {
+		t.Fatalf("no revoke: %+v", c)
+	}
+	_ = pressRun(tm, "y")
+	if got := toolsOf(); !slices.Equal(got, []string{"jira.*"}) {
+		t.Fatalf("after the revoke the team's grants are %v", got)
+	}
+}

internal/cli/tui_mirror.go

@@ -32,7 +32,7 @@ func (m consoleModel) mirrorClient() (*mirror.Client, error) {
 	if cfg.Mirror.API == "" {
 		return nil, errors.New("no mirror: mirror.api in server.yaml is empty")
 	}
-	return mirror.NewClient(cfg.Mirror.API, cfg.Mirror.TLS, 10*time.Second)
+	return mirror.NewFailoverClient(append([]string{cfg.Mirror.API}, cfg.Mirror.Peers...), cfg.Mirror.TLS, 10*time.Second)
 }
 
 func (m consoleModel) fetchMirror() tea.Cmd {

internal/config/config.go

@@ -345,6 +345,10 @@ type Mirror struct {
 	// listeners on the mirror precisely so this one can be locked down without
 	// re-certifying every agent's fetcher.
 	API string `koanf:"api"`
+	// Peers are the management addresses of mirrors kept in step with this
+	// one (llmgw mirror serve --peer): asked when it cannot be reached, and
+	// a node given the same file on one of them while it is down.
+	Peers []string `koanf:"peers"`
 	// TLS authenticates to the mirror's management API, with the same CA the
 	// agents use. Empty leaves the connection in plain HTTP, which is only
 	// acceptable on a network where nothing untrusted can reach that port —

internal/control/adminapi/api.go

@@ -205,6 +205,8 @@ func (a *API) Register(mux Mux) {
 	mux.HandleFunc("GET /api/v1/updates", a.listUpdates)
 	mux.HandleFunc("POST /api/v1/updates/check", a.checkUpdates)
 	mux.HandleFunc("POST /api/v1/models/{name}/update", a.updateModel)
+	mux.HandleFunc("POST /api/v1/models/{name}/context", a.setModelContext)
+	mux.HandleFunc("POST /api/v1/models/{name}/gpus", a.setModelGPUs)
 	mux.HandleFunc("POST /api/v1/models/{name}/rollback", a.rollbackModel)
 	a.registerCanary(mux)
 	a.registerEval(mux)
@@ -248,6 +250,9 @@ type NodeView struct {
 	MemoryGB int `json:"memory_gb,omitempty"`
 	// CPUCount is the machine's CPUs, 0 when the agent did not report them.
 	CPUCount int `json:"cpu_count,omitempty"`
+	// MemBandwidthGBs is how fast its memory reads, in GB/s, measured when
+	// its agent started; 0 when not reported. A model's speed on the CPU.
+	MemBandwidthGBs float64 `json:"mem_bandwidth_gbs,omitempty"`
 	// Engines are the inference engines installed here, as the agent found
 	// them. Absent means the agent is older than the probe and said nothing,
 	// which is not the same as having none — placement treats it that way too,
@@ -442,6 +447,10 @@ func loadView(l *inventory.Load, now time.Time) *LoadView {
 // ModelView is a model as `ctl models` shows it.
 type ModelView struct {
 	Name string `json:"name"`
+	// ContextLength is what one request may use, MaxOutput what it may
+	// answer; 0 when the catalog does not say.
+	ContextLength int `json:"context_length,omitempty"`
+	MaxOutput     int `json:"max_output,omitempty"`
 	// Canary is a share of its requests sent to a candidate; CanaryOf is
 	// set on the candidate, naming the model it is a canary of.
 	Canary   *catalog.Canary `json:"canary,omitempty"`
@@ -703,6 +712,15 @@ func (a *API) Nodes() []NodeView {
 				v.EngineBuilds[e] = b
 			}
 		}
+		v.MemBandwidthGBs = n.Host.MemBandwidthGBs
+		for _, g := range n.GPUs {
+			if g.MIG != nil {
+				continue // said of the card it is carved from
+			}
+			if p := g.PCIe.Problem(); p != "" {
+				v.Warnings = append(v.Warnings, fmt.Sprintf("GPU %d (%s) %s", g.Index, g.Name, p))
+			}
+		}
 		if h := n.Host; h.OSID != "" {
 			v.OS = strings.TrimSpace(h.OSID + " " + h.OSVersion + " " + h.Arch)
 			if h.Arch == "amd64" && !h.HasCPUFlag("avx2") {
@@ -825,6 +843,7 @@ func (a *API) Models() []ModelView {
 		h := health[m.Name]
 		v := ModelView{
 			Name: m.Name, Description: m.Description, Aliases: m.Aliases, Source: m.Source,
+			ContextLength: m.ContextLength, MaxOutput: m.MaxOutput,
 			Min: m.Scaling.Min, Max: m.Scaling.Max,
 			Ready: ready[m.Name], Total: total[m.Name],
 			Fallbacks: m.Fallbacks,
@@ -1736,6 +1755,46 @@ func (a *API) job(w http.ResponseWriter, r *http.Request) {
 	writeJSON(w, http.StatusOK, job)
 }
 
+// setModelContext changes a served model's context, its engines' arguments
+// and memory with it, and starts its replicas again on it.
+func (a *API) setModelContext(w http.ResponseWriter, r *http.Request) {
+	if a.serve == nil {
+		writeError(w, http.StatusNotImplemented, "this control plane cannot change a model's context")
+		return
+	}
+	var req modelserve.ContextRequest
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		writeError(w, http.StatusBadRequest, err.Error())
+		return
+	}
+	change, err := a.serve.SetContext(r.Context(), r.PathValue("name"), req, actorOf(r))
+	if err != nil {
+		writeError(w, http.StatusBadRequest, err.Error())
+		return
+	}
+	writeJSON(w, http.StatusOK, change)
+}
+
+// setModelGPUs spreads a served model over more cards, or fewer, and
+// starts its replicas again on it.
+func (a *API) setModelGPUs(w http.ResponseWriter, r *http.Request) {
+	if a.serve == nil {
+		writeError(w, http.StatusNotImplemented, "this control plane cannot change a model's cards")
+		return
+	}
+	var req modelserve.GPUsRequest
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		writeError(w, http.StatusBadRequest, err.Error())
+		return
+	}
+	change, err := a.serve.SetGPUs(r.Context(), r.PathValue("name"), req, actorOf(r))
+	if err != nil {
+		writeError(w, http.StatusBadRequest, err.Error())
+		return
+	}
+	writeJSON(w, http.StatusOK, change)
+}
+
 func (a *API) listUpdates(w http.ResponseWriter, _ *http.Request) {
 	out := make([]modelserve.UpdateInfo, 0)
 	if a.serve != nil {
@@ -1829,9 +1888,12 @@ func (a *API) setOrigin(w http.ResponseWriter, r *http.Request) {
 
 // PoolNodes is the fleet's ready nodes as a plan sees them: their cards,
 // which of them are free, the fabric they share, and their memory.
-func (a *API) PoolNodes() []modelplan.PoolNode {
+func (a *API) PoolNodes() []modelplan.PoolNode { return PoolNodesOf(a.Nodes()) }
+
+// PoolNodesOf is nodes as a plan sees them, the ready ones only.
+func PoolNodesOf(nodes []NodeView) []modelplan.PoolNode {
 	var out []modelplan.PoolNode
-	for _, n := range a.Nodes() {
+	for _, n := range nodes {
 		if n.State != "ready" {
 			continue
 		}

internal/control/adminapi/api_test.go

@@ -735,3 +735,17 @@ func TestRemovingANodeRevokesItsCertificate(t *testing.T) {
 		t.Error("the node was removed though its certificate was not revoked")
 	}
 }
+
+func TestANodesCardInTheWrongSlotIsAWarning(t *testing.T) {
+	n := gpuNode("g1", 2)
+	n.GPUs[0].PCIe = discovery.PCIeLink{Gen: 5, GPUGen: 5, Width: 16, WidthMax: 16}
+	n.GPUs[1].PCIe = discovery.PCIeLink{Gen: 5, GPUGen: 5, Width: 8, WidthMax: 16}
+	api, _ := setup(t, testModels(), n)
+	var warned []string
+	for _, v := range api.Nodes() {
+		warned = append(warned, v.Warnings...)
+	}
+	if len(warned) != 1 || !strings.Contains(warned[0], "GPU 1 (NVIDIA H100 80GB HBM3) linked at x8 of its x16 lanes") {
+		t.Errorf("warnings %q", warned)
+	}
+}

internal/control/adminapi/mcp.go

@@ -57,6 +57,9 @@ type MCPServerRequest struct {
 	// RatePerMinute bounds each caller's calls to its tools; a negative
 	// number takes the bound away.
 	RatePerMinute int `json:"rate_per_minute,omitempty"`
+	// Clear empties what an empty field above would leave as it is:
+	// "description", "timeout" (back to the default), "rate".
+	Clear []string `json:"clear,omitempty"`
 }
 
 // MCPTestView is what the server said when asked: its tools, or why not.
@@ -150,11 +153,18 @@ func (a *API) registerMCP(mux Mux) {
 // mcpServerFrom is the server a request makes, over the one of that name if
 // there is one: its secrets kept in the store, and named by reference.
 func (a *API) mcpServerFrom(ctx context.Context, name string, req MCPServerRequest, by string) (store.MCPServer, error) {
+	return MCPServerFrom(ctx, a.mcp, name, req, by)
+}
+
+// MCPServerFrom is the server a request makes, over the one of that name in
+// st if there is one: its secrets kept in the store, sealed, and named by
+// reference. Not stored; the caller puts it. For the API and the admin page.
+func MCPServerFrom(ctx context.Context, st MCPStore, name string, req MCPServerRequest, by string) (store.MCPServer, error) {
 	if !store.ValidMCPServerName(name) {
 		return store.MCPServer{}, errors.New("a server's name is lowercase letters, digits, - and _, starting with a letter: " +
 			"its tools are NAME.tool")
 	}
-	m, err := a.mcp.MCPServer(ctx, name)
+	m, err := st.MCPServer(ctx, name)
 	switch {
 	case errors.Is(err, store.ErrNotFound):
 		m = store.MCPServer{Name: name, Enabled: true, CreatedBy: by}
@@ -164,8 +174,8 @@ func (a *API) mcpServerFrom(ctx context.Context, name string, req MCPServerReque
 	if req.URL != "" {
 		m.URL = req.URL
 	}
-	if u, err := url.Parse(m.URL); m.URL != AdminMCPAddress && (err != nil || (u.Scheme != "https" && u.Scheme != "http") || u.Host == "") {
-		return m, errors.New("the server's url is an http:// or https:// address, or " + AdminMCPAddress + " for this cluster's own")
+	if err := CheckMCPAddress(m.URL); err != nil {
+		return m, err
 	}
 	if req.Description != "" {
 		m.Description = req.Description
@@ -184,7 +194,7 @@ func (a *API) mcpServerFrom(ctx context.Context, name string, req MCPServerReque
 			return m, errors.New("a secret header needs a name and a value")
 		}
 		key := "mcp/" + name + "/" + header
-		if err := a.mcp.SetClusterSecret(ctx, key, value); err != nil {
+		if err := st.SetClusterSecret(ctx, key, value); err != nil {
 			return m, err
 		}
 		m.HeaderRefs[header] = "store:" + key
@@ -202,12 +212,49 @@ func (a *API) mcpServerFrom(ctx context.Context, name string, req MCPServerReque
 	case req.RatePerMinute < 0:
 		m.RatePerMinute = 0
 	}
+	for _, f := range req.Clear {
+		switch f {
+		case "description":
+			m.Description = ""
+		case "timeout":
+			m.Timeout = 0
+		case "rate":
+			m.RatePerMinute = 0
+		default:
+			return m, errors.New("clear names description, timeout or rate, not " + f)
+		}
+	}
 	return m, nil
 }
 
+// CheckMCPAddress refuses an address a server cannot be reached at: an
+// http(s) one, or this cluster's own. Not one with a user and password in
+// it: a credential is a header, kept sealed, and an address is shown and
+// recorded in the audit log as written.
+func CheckMCPAddress(addr string) error {
+	if addr == AdminMCPAddress {
+		return nil
+	}
+	u, err := url.Parse(addr)
+	if err != nil || (u.Scheme != "https" && u.Scheme != "http") || u.Host == "" {
+		return errors.New("the server's url is an http:// or https:// address, or " + AdminMCPAddress + " for this cluster's own")
+	}
+	if u.User != nil {
+		return errors.New("the server's url has a user and password in it: give the credential as a header, which is kept sealed")
+	}
+	return nil
+}
+
 // testMCPServer asks a server for its tools, with its credentials, as /mcp
 // would.
 func (a *API) testMCPServer(ctx context.Context, m store.MCPServer) MCPTestView {
+	return TestMCPServer(ctx, m, http.HandlerFunc(a.MCP))
+}
+
+// TestMCPServer asks a server for its tools, with its credentials, as /mcp
+// would; local answers for this cluster's own (AdminMCPAddress), and
+// without it that one is not asked.
+func TestMCPServer(ctx context.Context, m store.MCPServer, local http.Handler) MCPTestView {
 	ctx, cancel := context.WithTimeout(ctx, 15*time.Second)
 	defer cancel()
 	c := &mcpclient.Client{URL: m.URL, Header: func(ctx context.Context) (http.Header, error) {
@@ -225,7 +272,10 @@ func (a *API) testMCPServer(ctx context.Context, m store.MCPServer) MCPTestView
 		return h, nil
 	}}
 	if m.URL == AdminMCPAddress {
-		c.URL, c.HTTP = InProcessURL, &http.Client{Transport: InProcess(http.HandlerFunc(a.MCP))}
+		if local == nil {
+			return MCPTestView{Error: "this cluster's own tools are answered only by a control plane"}
+		}
+		c.URL, c.HTTP = InProcessURL, &http.Client{Transport: InProcess(local)}
 	}
 	tools, err := c.Tools(ctx)
 	if err != nil {

internal/control/adminapi/routes.go

@@ -213,6 +213,11 @@ var Routes = []Route{
 	{Method: "POST", Path: "/api/v1/models/{name}/update", Summary: "Move a model to its repository's newer weights",
 		Description: "With a quality suite, or the model's standing gate, the new weights are asked it first as a hidden canary.",
 		Request:     modelserve.Change{}, OptionalBody: true, Response: modelserve.Job{}, Status: http.StatusAccepted},
+	{Method: "POST", Path: "/api/v1/models/{name}/context", Summary: "Change a served model's context: the catalog, its engines' arguments and memory, then its replicas restarted",
+		Request: modelserve.ContextRequest{}, Response: modelserve.ContextChange{}},
+	{Method: "POST", Path: "/api/v1/models/{name}/gpus", Summary: "Spread a served model over more cards or fewer: its profile, checked against its KV heads, its cards' memory and the nodes, then its replicas restarted",
+		Description: "Tensor parallel for vLLM, SGLang and TensorRT-LLM, a layer split for llama.cpp, on one node. With no room beside the running replicas, in_place stops each first.",
+		Request:     modelserve.GPUsRequest{}, Response: modelserve.GPUsChange{}},
 	{Method: "POST", Path: "/api/v1/models/{name}/rollback", Summary: "Go back to the weights the last update replaced",
 		Response: Result{}},
 	{Method: "POST", Path: "/api/v1/models/{name}/origin", Summary: "Record the repository a model's weights came from",

internal/control/agentsvc/convert.go

@@ -5,6 +5,8 @@ import (
 	"fmt"
 	"maps"
 	"runtime"
+	"slices"
+	"sync/atomic"
 	"time"
 
 	llmgwv1 "gitea.mynodes.xyz/ilker/llmgw/api/gen/llmgw/v1"
@@ -34,6 +36,7 @@ func NodeFromRegister(r *llmgwv1.Register) inventory.Node {
 		n.Host = discovery.Host{
 			OSID: h.GetOsId(), OSVersion: h.GetOsVersion(), OSCodename: h.GetOsCodename(),
 			Arch: h.GetArch(), CPUFlags: h.GetCpuFlags(), NVIDIADriver: h.GetNvidiaDriver(),
+			MemBandwidthGBs: h.GetMemBandwidthGbs(),
 		}
 	}
 	for _, w := range r.GetWeights() {
@@ -72,6 +75,8 @@ func NodeFromRegister(r *llmgwv1.Register) inventory.Node {
 				Problems:     g.GetProblems(),
 			},
 			MIG: migFrom(g.GetMig()),
+			PCIe: discovery.PCIeLink{Gen: int(g.GetPcieGen()), GPUGen: int(g.GetPcieGpuGen()),
+				Width: int(g.GetPcieWidth()), WidthMax: int(g.GetPcieWidthMax())},
 		})
 	}
 	n.IMEX = discovery.IMEX{State: r.GetImex().GetState(), Channels: int(r.GetImex().GetChannels())}
@@ -126,6 +131,10 @@ func RegisterFromInventory(nodeID, hostname, mgmtAddr, agentVersion string,
 			Problems:           g.Health.Problems,
 			TemperatureC:       int32(g.Health.TemperatureC),
 			EccErrors:          int32(g.Health.ECCErrors),
+			PcieGen:            int32(g.PCIe.Gen),
+			PcieGpuGen:         int32(g.PCIe.GPUGen),
+			PcieWidth:          int32(g.PCIe.Width),
+			PcieWidthMax:       int32(g.PCIe.WidthMax),
 		})
 		if f := g.Fabric; f != nil {
 			last := r.Gpus[len(r.Gpus)-1]
@@ -182,8 +191,31 @@ func AdoptedFromRegister(nodeID string, shards []*llmgwv1.RunningShard) []reconc
 }
 
 // StartShardFromCommand converts a reconciler command into a wire message.
+// sourceRewrite is where a model is fetched while its mirror is down: set
+// when mirrors are kept in step, else nil and the catalog's address as it is.
+var sourceRewrite atomic.Pointer[func(string) string]
+
+// SetSourceRewrite has every shard started from now on fetch its model, and
+// adapters, from the address f gives: one on a mirror that answers.
+func SetSourceRewrite(f func(string) string) { sourceRewrite.Store(&f) }
+
+func rewriteSource(src string) string {
+	if f := sourceRewrite.Load(); f != nil && src != "" {
+		return (*f)(src)
+	}
+	return src
+}
+
 func StartShardFromCommand(cmd reconciler.ShardCommand) *llmgwv1.StartShard {
 	p := cmd.Spec.Profile
+	p.ModelPath = rewriteSource(p.ModelPath)
+	if len(p.Adapters) > 0 {
+		ads := slices.Clone(p.Adapters)
+		for i := range ads {
+			ads[i].Path = rewriteSource(ads[i].Path)
+		}
+		p.Adapters = ads
+	}
 	return &llmgwv1.StartShard{
 		ReplicaId:     cmd.ReplicaID,
 		ShardId:       cmd.ShardID,
@@ -201,8 +233,9 @@ func StartShardFromCommand(cmd reconciler.ShardCommand) *llmgwv1.StartShard {
 			GpusPerNode: int32(p.GPUsPerNode), Nodes: int32(p.Nodes),
 			Args: p.Args, Env: p.Env,
 			ModelPath: p.ModelPath, ServedModelName: p.ServedModelName,
-			Adapters: adaptersToWire(p.Adapters),
-			KvEvents: p.KVEvents,
+			Adapters:    adaptersToWire(p.Adapters),
+			KvEvents:    p.KVEvents,
+			AllowPickle: p.AllowPickle,
 		},
 		StartupTimeoutMs: cmd.StartupTimeout.Milliseconds(),
 		Pool:             cmd.Pool,
@@ -238,8 +271,9 @@ func ShardSpecFromStart(m *llmgwv1.StartShard) (engines.ShardSpec, time.Duration
 			GPUsPerNode: int(p.GetGpusPerNode()), Nodes: int(p.GetNodes()),
 			Args: p.GetArgs(), Env: p.GetEnv(),
 			ModelPath: p.GetModelPath(), ServedModelName: p.GetServedModelName(),
-			Adapters: adaptersFromWire(p.GetAdapters()),
-			KVEvents: p.GetKvEvents(),
+			Adapters:    adaptersFromWire(p.GetAdapters()),
+			KVEvents:    p.GetKvEvents(),
+			AllowPickle: p.GetAllowPickle(),
 		},
 	}
 	return spec, time.Duration(m.GetStartupTimeoutMs()) * time.Millisecond

internal/control/agentsvc/convert_test.go

@@ -1,6 +1,7 @@
 package agentsvc
 
 import (
+	"strings"
 	"testing"
 
 	"gitea.mynodes.xyz/ilker/llmgw/internal/agent/engines"
@@ -18,3 +19,20 @@ func TestAdaptersTravelToTheAgentInOrder(t *testing.T) {
 		t.Errorf("adapters %+v", got)
 	}
 }
+
+// While its mirror is down, a shard is sent its model and adapters on the
+// mirror the rewrite names, and the command it was made from is left as it is.
+func TestAShardIsSentItsModelOnAMirrorThatAnswers(t *testing.T) {
+	SetSourceRewrite(func(s string) string { return strings.Replace(s, "http://mirror-a", "http://mirror-b", 1) })
+	t.Cleanup(func() { SetSourceRewrite(func(s string) string { return s }) })
+	ads := []engines.Adapter{{Name: "sql", Path: "http://mirror-a/sql.gguf"}}
+	cmd := reconciler.ShardCommand{ReplicaID: "r", ShardID: "s", Spec: engines.ShardSpec{
+		Profile: engines.Profile{Name: "cpu", Engine: "llamacpp", ModelPath: "http://mirror-a/qwen/v1/", Adapters: ads}}}
+	spec, _ := ShardSpecFromStart(StartShardFromCommand(cmd))
+	if spec.Profile.ModelPath != "http://mirror-b/qwen/v1/" || spec.Profile.Adapters[0].Path != "http://mirror-b/sql.gguf" {
+		t.Errorf("sent %+v", spec.Profile)
+	}
+	if ads[0].Path != "http://mirror-a/sql.gguf" {
+		t.Error("the catalog's adapter was changed in place")
+	}
+}

internal/control/agentsvc/heartbeat_test.go

@@ -325,3 +325,18 @@ func TestAProblemOnANodeWithoutGPUsDoesNotPanic(t *testing.T) {
 		t.Fatal("the node is gone")
 	}
 }
+
+func TestAHeartbeatCarriesTheMemorysMeasuredSpeed(t *testing.T) {
+	// Measured in the background after the agent starts, so it arrives in a
+	// heartbeat, not in the registration.
+	s := newTestService()
+	nodeWith(s, inventory.Node{ID: "n1", State: inventory.Ready, MemoryTotalMB: 7 * 1024})
+	s.applyHeartbeat("n1", &llmgwv1.Heartbeat{NodeId: "n1", MemBandwidthGbs: 41.9})
+	if n, _ := s.inv.Get("n1"); n.Host.MemBandwidthGBs != 41.9 {
+		t.Errorf("bandwidth %v, want 41.9", n.Host.MemBandwidthGBs)
+	}
+	s.applyHeartbeat("n1", &llmgwv1.Heartbeat{NodeId: "n1"})
+	if n, _ := s.inv.Get("n1"); n.Host.MemBandwidthGBs != 41.9 {
+		t.Errorf("a heartbeat without it erased it: %v", n.Host.MemBandwidthGBs)
+	}
+}

internal/control/agentsvc/service.go

@@ -547,10 +547,11 @@ func (s *Service) applyHeartbeat(nodeID string, hb *llmgwv1.Heartbeat) {
 	storage := hb.GetStorage()
 	memMB := int(hb.GetMemoryTotalMb())
 	cpus := int(hb.GetCpuCount())
+	bw := hb.GetMemBandwidthGbs()
 	if len(used) == 0 && len(foreign) == 0 && len(problems) == 0 && len(storage) == 0 &&
 		!hb.GetWeightsChanged() && memMB == 0 && cpus == 0 && !hb.GetEnginesChanged() &&
 		!hb.GetProvisionChanged() && hb.GetLoad() == nil && len(hb.GetIbPorts()) == 0 &&
-		!hb.GetFabricsChanged() && len(hb.GetGpuFabric()) == 0 && hb.GetImex() == nil {
+		!hb.GetFabricsChanged() && len(hb.GetGpuFabric()) == 0 && hb.GetImex() == nil && bw == 0 {
 		return
 	}
 
@@ -622,6 +623,9 @@ func (s *Service) applyHeartbeat(nodeID string, hb *llmgwv1.Heartbeat) {
 			"from_mb", n.MemoryTotalMB, "to_mb", memMB)
 		n.MemoryTotalMB = memMB
 	}
+	if bw > 0 {
+		n.Host.MemBandwidthGBs = bw
+	}
 	if cpus > 0 && cpus != n.CPUCount {
 		if n.CPUCount > 0 {
 			s.log.Info("node CPUs changed", "node", nodeID, "from", n.CPUCount, "to", cpus)

internal/control/catalog/catalog.go

@@ -323,9 +323,15 @@ type Model struct {
 	// choice by guesswork.
 	Description string `json:"description,omitempty" koanf:"description"`
 	// Source is the weights path or a HuggingFace id.
-	Source   string    `json:"source,omitempty" koanf:"source"`
-	Profiles []Profile `json:"profiles,omitempty" koanf:"profiles"`
-	Scaling  Scaling   `json:"scaling,omitempty" koanf:"scaling"`
+	Source string `json:"source,omitempty" koanf:"source"`
+	// AllowPickle lets the nodes fetch this model's weights, or its
+	// adapters', in a pickle format (.bin, .pt, .pth, .ckpt), which run code
+	// as they are loaded. Off, such a source is refused when the model is
+	// added and by the node before it fetches anything. For a model that has
+	// no safetensors or GGUF form and whose files were checked.
+	AllowPickle bool      `json:"allow_pickle,omitempty" koanf:"allow_pickle"`
+	Profiles    []Profile `json:"profiles,omitempty" koanf:"profiles"`
+	Scaling     Scaling   `json:"scaling,omitempty" koanf:"scaling"`
 	// Class is how much this model's capacity is worth against another's,
 	// when there is not enough for both: critical, standard or opportunistic.
 	// Empty is standard, or opportunistic for a model loaded on demand. See
@@ -1185,6 +1191,22 @@ func (m Model) validate() error {
 			}
 			continue
 		}
+		// Strata prepares one of the few families it runs from their GGUF
+		// files, on one card, and has to be told which.
+		if p.Engine == "strata" {
+			if p.Nodes > 1 || p.TP > 1 || p.PP > 1 || p.GPUsPerNode > 1 {
+				return fmt.Errorf("model %q profile %q: strata runs a replica on one GPU of one node", m.Name, p.Name)
+			}
+			if !slices.ContainsFunc(p.Args, func(a string) bool { return a == "--family" || strings.HasPrefix(a, "--family=") }) {
+				return fmt.Errorf("model %q profile %q: strata needs the model's family and size in args "+
+					"(--family qwen --size IQ2_XS)", m.Name, p.Name)
+			}
+			if strings.HasSuffix(strings.ToLower(strings.TrimSuffix(m.SourceOf(p), "/")), ".gguf") {
+				return fmt.Errorf("model %q profile %q: strata prepares a model from all of its GGUF files; "+
+					"give the directory they are in as the source", m.Name, p.Name)
+			}
+			continue
+		}
 		// TensorRT-LLM spans machines only under MPI started from one of them,
 		// which the agents, each starting its own node, do not do; and it
 		// loads a Hugging Face checkpoint, not llama.cpp's format.
@@ -1336,10 +1358,16 @@ func (m Model) EngineProfile(name string) (engines.Profile, bool) {
 				Name   string `json:"name"`
 				Source string `json:"source"`
 				HFRepo string `json:"hf_repo,omitempty"`
+				// A string: the node reads the members as strings.
+				AllowPickle string `json:"allow_pickle,omitempty"`
 			}
 			var ms []member
 			for _, sm := range m.Shared.Members {
-				ms = append(ms, member{Name: sm.Name, Source: sm.Source, HFRepo: sm.HFRepo})
+				mb := member{Name: sm.Name, Source: sm.Source, HFRepo: sm.HFRepo}
+				if sm.AllowPickle {
+					mb.AllowPickle = "true"
+				}
+				ms = append(ms, mb)
 			}
 			b, _ := json.Marshal(ms)
 			env = maps.Clone(env)
@@ -1369,6 +1397,7 @@ func (m Model) EngineProfile(name string) (engines.Profile, bool) {
 			ServedModelName: m.Name,
 			Adapters:        m.engineAdapters(),
 			KVEvents:        p.KVEvents,
+			AllowPickle:     m.AllowPickle,
 		}.WithRuntimeLoRA(m.MaxLoadedAdapters), true
 	}
 	return engines.Profile{}, false

internal/control/catalog/pickle.go

@@ -0,0 +1,91 @@
+package catalog
+
+import (
+	"fmt"
+	"path"
+	"slices"
+	"strings"
+
+	"gitea.mynodes.xyz/ilker/llmgw/internal/agent/engines"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/agent/weights"
+)
+
+// PickleSources are the sources of a model's weights and adapters that are
+// in a pickle format, which run code as they are loaded (weights.Pickled).
+func (m Model) PickleSources() []string {
+	var out []string
+	add := func(s string) {
+		if s != "" && weights.Pickled(s) && !slices.Contains(out, s) {
+			out = append(out, s)
+		}
+	}
+	add(m.Source)
+	for _, p := range m.Profiles {
+		add(p.Source)
+	}
+	for _, a := range m.Adapters {
+		add(a.Source)
+	}
+	return out
+}
+
+// CheckPickle refuses a model whose weights are in a pickle format, unless
+// it says allow_pickle. Asked when a model is added or its sources change,
+// not when the catalog is built: a model already running is refused by its
+// node, with the same words, rather than dropped from the catalog.
+// Sources in except were there before and are not asked about again.
+func (m Model) CheckPickle(except ...string) error {
+	if m.AllowPickle {
+		return nil
+	}
+	for _, s := range m.PickleSources() {
+		if slices.Contains(except, s) {
+			continue
+		}
+		return fmt.Errorf("model %q: %s is in a pickle format, which runs code when it is loaded; "+
+			"use the model's safetensors or GGUF, or set allow_pickle: true if it was checked and has no other form",
+			m.Name, s)
+	}
+	// What an engine downloads itself, by a Hugging Face id, no node looks
+	// at: vLLM reads a model by id as safetensors only, but not an adapter,
+	// and TensorRT-LLM neither.
+	for _, s := range m.UnlookedSources() {
+		if slices.Contains(except, s) {
+			continue
+		}
+		return fmt.Errorf("model %q: %s is a Hugging Face id its engine downloads itself, and may load in a pickle "+
+			"format, which runs code; mirror its safetensors and give that, or set allow_pickle: true if it was checked",
+			m.Name, s)
+	}
+	return nil
+}
+
+// UnlookedSources are those an engine fetches by a Hugging Face id, past
+// the node's check: adapters, and a TensorRT-LLM model's.
+func (m Model) UnlookedSources() []string {
+	var out []string
+	for _, a := range m.Adapters {
+		if hubID(a.Source) {
+			out = append(out, a.Source)
+		}
+	}
+	for _, p := range m.Profiles {
+		if p.Engine == "trtllm" {
+			if src := m.SourceOf(p); hubID(src) && !slices.Contains(out, src) {
+				out = append(out, src)
+			}
+		}
+	}
+	return out
+}
+
+// hubID reports whether a source is a Hugging Face id, as the agent reads
+// one (engines.IsHubID) — gpt2 too — and not a file named so, nor a GGUF
+// in a repository (org/name:Q4_K_M).
+func hubID(s string) bool {
+	if !engines.IsHubID(s) || strings.Contains(s, ":") {
+		return false
+	}
+	ext := strings.ToLower(path.Ext(s))
+	return ext != ".gguf" && ext != ".safetensors" && !weights.Pickled(s)
+}

internal/control/catalog/pickle_test.go

@@ -0,0 +1,76 @@
+package catalog
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestAModelInAPickleFormatIsRefusedUnlessItSaysSo(t *testing.T) {
+	m := Model{Name: "triposr", Source: "hf:stabilityai/TripoSR/model.ckpt",
+		Adapters: []Adapter{{Name: "a", Source: "http://m/a/adapter_model.bin"}}}
+	err := m.CheckPickle()
+	if err == nil || !strings.Contains(err.Error(), "model.ckpt") || !strings.Contains(err.Error(), "allow_pickle") {
+		t.Fatalf("not refused, or not said why: %v", err)
+	}
+	if got := m.PickleSources(); len(got) != 2 {
+		t.Errorf("pickle sources %v, want the model's and the adapter's", got)
+	}
+	// A source that was there before is not asked about again: changing
+	// the scaling of a model that runs does not need allow_pickle.
+	if err := m.CheckPickle(m.PickleSources()...); err != nil {
+		t.Errorf("sources already there were refused: %v", err)
+	}
+	m.AllowPickle = true
+	if err := m.CheckPickle(); err != nil {
+		t.Errorf("allow_pickle was not honoured: %v", err)
+	}
+	if err := (Model{Name: "q", Source: "http://m/q/v/q-Q4_K_M.gguf"}).CheckPickle(); err != nil {
+		t.Errorf("a GGUF was refused: %v", err)
+	}
+}
+
+func TestTheTripoSRPresetSaysItsCheckpointIsAllowed(t *testing.T) {
+	p, ok := PresetNamed("triposr")
+	if !ok {
+		t.Fatal("no triposr preset")
+	}
+	if err := p.Model.CheckPickle(); err != nil {
+		t.Errorf("the preset llmgw ships is refused by its own check: %v", err)
+	}
+}
+
+func TestWhatAnEngineDownloadsByIDIsAskedAbout(t *testing.T) {
+	m := Model{Name: "m", Source: "http://mirror/m/v/", Profiles: []Profile{{Name: "gpu", Engine: "vllm"}},
+		Adapters: []Adapter{{Name: "a", Source: "someone/lora-x"}}}
+	if err := m.CheckPickle(); err == nil || !strings.Contains(err.Error(), "someone/lora-x") {
+		t.Errorf("an adapter by its id: %v", err)
+	}
+	m.Adapters = []Adapter{{Name: "a", Source: "http://mirror/a/v/adapter_model.safetensors"}}
+	if err := m.CheckPickle(); err != nil {
+		t.Errorf("a mirrored adapter: %v", err)
+	}
+	trt := Model{Name: "t", Source: "meta-llama/Llama-3.1-8B-Instruct", Profiles: []Profile{{Name: "gpu", Engine: "trtllm"}}}
+	if err := trt.CheckPickle(); err == nil {
+		t.Error("TensorRT-LLM downloading by id was not asked about")
+	}
+	vllm := Model{Name: "v", Source: "meta-llama/Llama-3.1-8B-Instruct", Profiles: []Profile{{Name: "gpu", Engine: "vllm"}}}
+	if err := vllm.CheckPickle(); err != nil {
+		t.Errorf("vLLM reads a model by id as safetensors only: %v", err)
+	}
+	for s, want := range map[string]bool{"Qwen/Qwen2.5-7B": true, "meta-llama/Llama-3.1-8B-Instruct": true,
+		"org/model.gguf": false, "o/r:Q4_K_M": false, "/models/x": false, "hf:o/r/f.gguf": false,
+		"gpt2": true, "~me/x": false} {
+		if hubID(s) != want {
+			t.Errorf("hubID(%q) = %v", s, !want)
+		}
+	}
+}
+
+func TestEveryPresetPassesItsOwnPickleCheck(t *testing.T) {
+	// What llmgw ships must be addable: TripoSR was not, once.
+	for _, p := range Presets() {
+		if err := p.Model.CheckPickle(); err != nil {
+			t.Errorf("%s: %v", p.Model.Name, err)
+		}
+	}
+}

internal/control/catalog/presets.go

@@ -52,6 +52,9 @@ func Presets() []Preset {
 				Name:        "triposr",
 				Description: "TripoSR: a 3D model (GLB) from a picture of one object on a plain background, as a job",
 				Source:      "hf:stabilityai/TripoSR/model.ckpt",
+				// TripoSR is published as this checkpoint only; it is
+				// Stability AI's, pinned, and the one pickle llmgw ships.
+				AllowPickle: true,
 				ComfyUI: &ComfyUI{
 					Workflow: "triposr",
 					Output:   OutputMeshes,

internal/control/catalog/shared.go

@@ -34,6 +34,8 @@ type SharedMember struct {
 	Source string `json:"source"`
 	// HFRepo is the repository its node loads the weights by name from.
 	HFRepo string `json:"hf_repo,omitempty"`
+	// AllowPickle is the member model's allow_pickle.
+	AllowPickle bool `json:"allow_pickle,omitempty"`
 	// Workflow and Checkpoint are what a request for it is sent.
 	Workflow   string `json:"workflow"`
 	Checkpoint string `json:"checkpoint"`
@@ -80,7 +82,7 @@ func envModel(env Environment, members []Model) Model {
 			hf = m.ComfyUI.HuggingFaceRepo
 		}
 		shared.Members = append(shared.Members, SharedMember{Name: m.Name, Aliases: m.Aliases,
-			Source: m.Source, HFRepo: hf, Workflow: workflow, Checkpoint: ckpt})
+			Source: m.Source, HFRepo: hf, AllowPickle: m.AllowPickle, Workflow: workflow, Checkpoint: ckpt})
 		idle = max(idle, m.Scaling.IdleTimeout)
 		minN, maxN = max(minN, m.Scaling.Min), max(maxN, m.Scaling.Max)
 		if ClassRank(m.EffectiveClass()) > ClassRank(class) || class == "" {
@@ -97,6 +99,7 @@ func envModel(env Environment, members []Model) Model {
 		Role:        RoleEnvironment,
 		Description: "the shared ComfyUI replica of environment " + env.Name,
 		Source:      first.Source,
+		AllowPickle: first.AllowPickle,
 		ComfyUI:     &ComfyUI{Environment: env.Name, Nodes: env.Nodes, Output: OutputEnvironment},
 		Profiles:    profs,
 		Scaling:     Scaling{Min: minN, Max: maxN, IdleTimeout: idle},

internal/control/catalog/strata_test.go

@@ -0,0 +1,31 @@
+package catalog
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestAStrataProfileNamesItsFamilyAndRunsOnOneCard(t *testing.T) {
+	ok := Model{Name: "q38", Source: "http://mirror:8081/q38/ab12cd34/", Profiles: []Profile{{
+		Name: "p", Engine: "strata", GPUClass: "6000-24g",
+		Args: []string{"--family", "qwen", "--size", "IQ2_XS", "--context", "32768"},
+	}}}
+	if _, err := New([]Model{ok}); err != nil {
+		t.Fatalf("a one-card Strata profile was refused: %v", err)
+	}
+	for want, bad := range map[string]Model{
+		"one GPU of one node": {Name: "x", Source: "http://mirror:8081/x/1/", Profiles: []Profile{{
+			Name: "p", Engine: "strata", GPUClass: "6000-24g", GPUsPerNode: 2, TP: 2,
+			Args: []string{"--family", "qwen"}}}},
+		"family and size": {Name: "x", Source: "http://mirror:8081/x/1/", Profiles: []Profile{{
+			Name: "p", Engine: "strata", GPUClass: "6000-24g", Args: []string{"--size", "IQ2_XS"}}}},
+		"all of its GGUF files": {Name: "x", Source: "http://mirror:8081/x/1/x-00001-of-00002.gguf", Profiles: []Profile{{
+			Name: "p", Engine: "strata", GPUClass: "6000-24g", Args: []string{"--family", "qwen"}}}},
+		"--gguf-dir": {Name: "x", Source: "http://mirror:8081/x/1/", Profiles: []Profile{{
+			Name: "p", Engine: "strata", GPUClass: "6000-24g", Args: []string{"--family", "qwen", "--gguf-dir", "/etc"}}}},
+	} {
+		if _, err := New([]Model{bad}); err == nil || !strings.Contains(err.Error(), want) {
+			t.Errorf("expected %q, got %v", want, err)
+		}
+	}
+}

internal/control/modelserve/context.go

@@ -0,0 +1,233 @@
+package modelserve
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"math"
+	"slices"
+	"strconv"
+	"strings"
+
+	"gitea.mynodes.xyz/ilker/llmgw/internal/control/catalog"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/gguf"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/modelplan"
+)
+
+// A model's context changed after it was served: what one request may use,
+// in the three places it lives — the catalog, the engine's arguments, and
+// the memory its KV cache takes — together, checked against what the cards
+// hold, then its replicas started again on it one by one.
+
+// ContextRequest is the context to change to: tokens one request may use,
+// and for llama.cpp how many at once (its pool is the two multiplied).
+type ContextRequest struct {
+	Tokens int `json:"tokens"`
+	Slots  int `json:"slots,omitempty"`
+}
+
+// ContextProfile is what one profile needs before and after.
+type ContextProfile struct {
+	Profile string   `json:"profile"`
+	Engine  string   `json:"engine"`
+	Before  string   `json:"before,omitempty"`
+	After   string   `json:"after,omitempty"`
+	Args    []string `json:"args"`
+}
+
+// ContextChange is what was changed, and the replicas started again on it.
+type ContextChange struct {
+	Model     string           `json:"model"`
+	Before    int              `json:"before"`
+	After     int              `json:"after"`
+	MaxOutput int              `json:"max_output"`
+	Trained   int              `json:"trained,omitempty"`
+	Profiles  []ContextProfile `json:"profiles"`
+	Restarted int              `json:"restarted"`
+	// Note is what could not be checked, said once.
+	Note string `json:"note,omitempty"`
+}
+
+// shape is what sizes a context: a token's KV cache, the context the model
+// was trained for, and its weights; and its KV heads, which decide how many
+// cards tensor parallel can spread it over (0 unknown).
+type shape struct {
+	kvPerToken int64
+	trained    int
+	weights    int64
+	kvHeads    int
+}
+
+// SetContext changes a served model's context.
+func (s *Server) SetContext(ctx context.Context, name string, req ContextRequest, by string) (ContextChange, error) {
+	if s.deps.Current == nil || s.deps.Models == nil {
+		return ContextChange{}, errors.New("this control plane cannot change the catalog")
+	}
+	var m catalog.Model
+	found := false
+	for _, c := range s.deps.Current() {
+		if c.Name == name {
+			m, found = c, true
+		}
+	}
+	switch {
+	case !found:
+		return ContextChange{}, fmt.Errorf("no model %q in the catalog", name)
+	case s.deps.Models.Declared(name):
+		return ContextChange{}, fmt.Errorf("model %q is declared in the cluster's settings: change its context there (llmgw ctl config edit)", name)
+	case req.Tokens < 256:
+		return ContextChange{}, errors.New("a context of at least 256 tokens")
+	case req.Slots < 0 || req.Slots > 64:
+		return ContextChange{}, errors.New("slots are 1 to 64")
+	}
+	sh, note := s.shapeOf(ctx, m)
+	if sh.trained > 0 && req.Tokens > sh.trained {
+		return ContextChange{}, fmt.Errorf("%s was trained for %d tokens; a longer context answers worse, and llama.cpp or vLLM refuses it",
+			name, sh.trained)
+	}
+	out := ContextChange{Model: name, Before: m.ContextLength, After: req.Tokens, Trained: sh.trained, Note: note}
+	profiles := slices.Clone(m.Profiles)
+	for i, p := range profiles {
+		next, change, err := resizeProfile(p, req, sh)
+		if err != nil {
+			return ContextChange{}, err
+		}
+		profiles[i] = next
+		out.Profiles = append(out.Profiles, change)
+	}
+	out.MaxOutput = m.MaxOutput
+	if m.MaxOutput == 0 || m.MaxOutput == maxOutput(m.ContextLength) {
+		out.MaxOutput = maxOutput(req.Tokens)
+	}
+	out.MaxOutput = min(out.MaxOutput, req.Tokens)
+	if _, err := s.deps.Models.SetContext(ctx, name, req.Tokens, out.MaxOutput, profiles, by); err != nil {
+		return ContextChange{}, err
+	}
+	// Started again on it, a new replica before the old one stops.
+	var ids []string
+	if s.deps.Fleet != nil {
+		for _, r := range s.deps.Fleet.Replicas() {
+			if r.Model == name {
+				ids = append(ids, r.ID)
+			}
+		}
+	}
+	if len(ids) > 0 && s.deps.Restart != nil {
+		if err := s.deps.Restart(ids...); err != nil {
+			return out, fmt.Errorf("the context is changed, and its replicas could not be started again on it: %w", err)
+		}
+		out.Restarted = len(ids)
+	}
+	return out, nil
+}
+
+// shapeOf reads what sizes a model's context: a GGUF file's header from the
+// mirror, else the original weights' config from their repository.
+func (s *Server) shapeOf(ctx context.Context, m catalog.Model) (shape, string) {
+	var sh shape
+	if m.Origin != nil {
+		for _, f := range m.Origin.Files {
+			sh.weights += f.Bytes
+		}
+	}
+	switch {
+	case strings.HasSuffix(strings.ToLower(m.Source), ".gguf"):
+		h, err := gguf.FromURL(ctx, nil, m.Source)
+		if err != nil {
+			return sh, "its header could not be read (" + err.Error() + "), so the memory it needs is not checked"
+		}
+		sh.kvPerToken, sh.trained = h.KVBytesPerToken(), int(h.ContextLength) //nolint:gosec // a header's context
+		sh.kvHeads = int(h.HeadCountKV)                                       //nolint:gosec // a header's head count
+	case m.Origin != nil && m.Origin.Repo != "":
+		mm, err := modelplan.FetchModel(ctx, m.Origin.Repo)
+		if err != nil {
+			return sh, "its config could not be read (" + err.Error() + "), so the memory it needs is not checked"
+		}
+		sh.kvPerToken, sh.trained = mm.KVBytesPerToken(modelplan.BF16), mm.MaxContext
+		sh.kvHeads = mm.KVHeads
+	default:
+		return sh, "where it came from is not known, so the memory it needs is not checked"
+	}
+	if sh.weights == 0 {
+		return sh, "its files' sizes are not known, so the memory it needs is not checked"
+	}
+	return sh, ""
+}
+
+// resizeProfile is a profile with the context asked for: its engine's
+// arguments, and its memory where it is on CPUs, refused where its cards
+// would not hold it.
+func resizeProfile(p catalog.Profile, req ContextRequest, sh shape) (catalog.Profile, ContextProfile, error) {
+	change := ContextProfile{Profile: p.Name, Engine: p.Engine}
+	var pool int
+	switch p.Engine {
+	case "llamacpp":
+		slots := req.Slots
+		if slots == 0 {
+			slots = max(argInt(p.Args, "--parallel"), 1)
+		}
+		pool = req.Tokens * slots
+		p.Args = setArg(setArg(p.Args, "--ctx-size", strconv.Itoa(pool)), "--parallel", strconv.Itoa(slots))
+	case "vllm":
+		pool = req.Tokens
+		p.Args = setArg(p.Args, "--max-model-len", strconv.Itoa(pool))
+	case "sglang":
+		pool = req.Tokens
+		p.Args = setArg(p.Args, "--context-length", strconv.Itoa(pool))
+	default:
+		return p, change, fmt.Errorf("a %s profile (%s) does not take a context this way", p.Engine, p.Name)
+	}
+	change.Args = p.Args
+	if sh.kvPerToken == 0 || sh.weights == 0 {
+		return p, change, nil
+	}
+	if p.CPUOnly() {
+		need := sh.weights + int64(pool)*sh.kvPerToken + int64(1<<30) + sh.weights/20
+		gb := int(math.Ceil(modelplan.GB(need)))
+		change.Before, change.After = fmt.Sprintf("%d GB of RAM", p.MinMemoryGB), fmt.Sprintf("%d GB of RAM", gb)
+		p.MinMemoryGB = gb
+		return p, change, nil
+	}
+	cards := max(p.GPUsPerNode, p.TP, 1) * max(p.Nodes, p.PP, 1)
+	need := modelplan.CardNeed(p.Engine, sh.weights, int64(pool)*sh.kvPerToken, cards)
+	change.After = fmt.Sprintf("%.1f GB a card of %d", need, cards)
+	if budget := modelplan.CardBudget(p.GPUClass); budget > 0 {
+		change.After += fmt.Sprintf(", of %.0f usable", budget)
+		if need > budget {
+			return p, change, fmt.Errorf("with %d tokens the %s profile needs %.1f GB on each of its %d %s cards, more than the %.0f GB they hold: "+
+				"a shorter context, or the model on more cards", req.Tokens, p.Name, need, cards, p.GPUClass, budget)
+		}
+	}
+	return p, change, nil
+}
+
+// setArg sets a flag's value in arguments, or adds it.
+func setArg(args []string, flag, value string) []string {
+	out := slices.Clone(args)
+	for i := 0; i < len(out); i++ {
+		if out[i] == flag && i+1 < len(out) {
+			out[i+1] = value
+			return out
+		}
+		if v, ok := strings.CutPrefix(out[i], flag+"="); ok && v != "" {
+			out[i] = flag + "=" + value
+			return out
+		}
+	}
+	return append(out, flag, value)
+}
+
+// argInt is a flag's whole-number value, 0 when absent.
+func argInt(args []string, flag string) int {
+	for i, a := range args {
+		if a == flag && i+1 < len(args) {
+			n, _ := strconv.Atoi(args[i+1])
+			return n
+		}
+		if v, ok := strings.CutPrefix(a, flag+"="); ok {
+			n, _ := strconv.Atoi(v)
+			return n
+		}
+	}
+	return 0
+}

internal/control/modelserve/context_test.go

@@ -0,0 +1,90 @@
+package modelserve
+
+import (
+	"context"
+	"slices"
+	"strings"
+	"testing"
+
+	"gitea.mynodes.xyz/ilker/llmgw/internal/control/catalog"
+)
+
+func TestAContextChangeResizesTheEngineAndTheMemoryAndRestartsTheReplicas(t *testing.T) {
+	s, _, models, _, fleet := updSetup(t)
+	m := models.models["qwen"]
+	m.ContextLength, m.MaxOutput = 4096, maxOutput(4096)
+	m.Origin = &catalog.Origin{Repo: "o/r", Files: []catalog.OriginFile{{Path: "q4.gguf", Bytes: 2 << 30}}}
+	m.Profiles[0].Args = []string{"--ctx-size", "8192", "--parallel", "2"}
+	models.models["qwen"] = m
+
+	got, err := s.SetContext(context.Background(), "qwen", ContextRequest{Tokens: 32768}, "admin")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got.Before != 4096 || got.After != 32768 || got.Restarted != 2 || got.Note != "" {
+		t.Fatalf("%+v", got)
+	}
+	after := models.models["qwen"]
+	if after.ContextLength != 32768 || after.MaxOutput != maxOutput(32768) {
+		t.Errorf("the catalog says %d tokens, %d out", after.ContextLength, after.MaxOutput)
+	}
+	// Two slots kept: the pool is both.
+	if a := after.Profiles[0].Args; !slices.Equal(a, []string{"--ctx-size", "65536", "--parallel", "2"}) {
+		t.Errorf("args %v", a)
+	}
+	// 2 GiB of weights, 64k tokens at 28 KiB each, an engine's share.
+	if gb := after.Profiles[0].MinMemoryGB; gb < 4 || gb > 6 {
+		t.Errorf("memory %d GB", gb)
+	}
+	for _, r := range fleet.Replicas() {
+		if !strings.Contains(r.ID, "-new") {
+			t.Errorf("%s was not started again", r.ID)
+		}
+	}
+}
+
+func TestAContextTheCardsCannotHoldIsRefusedWithTheFigures(t *testing.T) {
+	sh := shape{kvPerToken: 128 << 10, weights: 8 << 30}
+	p := catalog.Profile{Name: "gpu", Engine: "vllm", GPUClass: "t4-16g", GPUsPerNode: 1}
+	if _, _, err := resizeProfile(p, ContextRequest{Tokens: 131072}, sh); err == nil || !strings.Contains(err.Error(), "GB") {
+		t.Fatalf("not refused: %v", err)
+	}
+	next, change, err := resizeProfile(p, ContextRequest{Tokens: 4096}, sh)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !slices.Equal(next.Args, []string{"--max-model-len", "4096"}) || !strings.Contains(change.After, "a card of 1") {
+		t.Errorf("%v %+v", next.Args, change)
+	}
+}
+
+func TestAContextIsRefusedForADeclaredModelOrPastWhatItWasTrainedFor(t *testing.T) {
+	s, _, models, _, _ := updSetup(t)
+	models.declared["qwen"] = true
+	if _, err := s.SetContext(context.Background(), "qwen", ContextRequest{Tokens: 8192}, "admin"); err == nil {
+		t.Error("a declared model's context was changed")
+	}
+	models.declared["qwen"] = false
+	if _, err := s.SetContext(context.Background(), "qwen", ContextRequest{Tokens: 100}, "admin"); err == nil {
+		t.Error("100 tokens taken")
+	}
+	if _, err := s.SetContext(context.Background(), "nope", ContextRequest{Tokens: 8192}, "admin"); err == nil {
+		t.Error("no such model, and no error")
+	}
+}
+
+func TestEngineArgumentsAreSetInEitherSpelling(t *testing.T) {
+	if a := setArg([]string{"--ctx-size=4096", "-ngl", "99"}, "--ctx-size", "8192"); !slices.Equal(a, []string{"--ctx-size=8192", "-ngl", "99"}) {
+		t.Error(a)
+	}
+	if a := setArg([]string{"-ngl", "99"}, "--parallel", "4"); !slices.Equal(a, []string{"-ngl", "99", "--parallel", "4"}) {
+		t.Error(a)
+	}
+	if argInt([]string{"--parallel=3"}, "--parallel") != 3 || argInt(nil, "--parallel") != 0 {
+		t.Error("argInt")
+	}
+	p := catalog.Profile{Name: "x", Engine: "comfyui"}
+	if _, _, err := resizeProfile(p, ContextRequest{Tokens: 4096}, shape{}); err == nil {
+		t.Error("a comfyui profile took a context")
+	}
+}

internal/control/modelserve/gpus.go

@@ -0,0 +1,285 @@
+package modelserve
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"slices"
+	"strings"
+
+	"gitea.mynodes.xyz/ilker/llmgw/internal/control/catalog"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/modelplan"
+)
+
+// A served model moved to more cards, or fewer: how many each replica
+// spreads it over, on one node. The profile changes — tensor parallel for
+// vLLM, SGLang and TensorRT-LLM, llama.cpp's layer split for llama.cpp —
+// checked against the model's KV heads, what each card would hold, and the
+// nodes there are; then its replicas start again on it.
+
+// GPUsRequest is how many cards a replica is to use.
+type GPUsRequest struct {
+	GPUs int `json:"gpus"`
+	// Profile is the one to change, when the model has more than one
+	// profile on cards.
+	Profile string `json:"profile,omitempty"`
+	// InPlace stops each replica before its replacement starts: for when no
+	// node has the cards free beside it. It does not answer while the new
+	// one loads.
+	InPlace bool `json:"in_place,omitempty"`
+}
+
+// GPUsChange is what was changed, and the replicas started again on it.
+type GPUsChange struct {
+	Model   string   `json:"model"`
+	Profile string   `json:"profile"`
+	Engine  string   `json:"engine"`
+	Class   string   `json:"gpu_class"`
+	Before  int      `json:"before"`
+	After   int      `json:"after"`
+	Args    []string `json:"args"`
+	// PerCard is what each card holds after, when it could be worked out.
+	PerCard   string `json:"per_card,omitempty"`
+	Restarted int    `json:"restarted"`
+	InPlace   bool   `json:"in_place,omitempty"`
+	// Note is what could not be checked, said once.
+	Note string `json:"note,omitempty"`
+}
+
+// tpFlags are each engine's own spelling of a tensor-parallel width, which
+// the agent sets from the profile: a profile's own would contradict it.
+var tpFlags = map[string][]string{
+	"vllm":   {"--tensor-parallel-size", "-tp"},
+	"sglang": {"--tp-size", "--tensor-parallel-size", "--tp"},
+	"trtllm": {"--tp_size"},
+	// llama.cpp's proportions were for the cards it had.
+	"llamacpp": {"--tensor-split", "-ts"},
+}
+
+// SetGPUs changes how many cards a served model's replicas use.
+func (s *Server) SetGPUs(ctx context.Context, name string, req GPUsRequest, by string) (GPUsChange, error) {
+	if s.deps.Current == nil || s.deps.Models == nil {
+		return GPUsChange{}, errors.New("this control plane cannot change the catalog")
+	}
+	var m catalog.Model
+	found := false
+	for _, c := range s.deps.Current() {
+		if c.Name == name {
+			m, found = c, true
+		}
+	}
+	switch {
+	case !found:
+		return GPUsChange{}, fmt.Errorf("no model %q in the catalog", name)
+	case s.deps.Models.Declared(name):
+		return GPUsChange{}, fmt.Errorf("model %q is declared in the cluster's settings: change its profile there (llmgw ctl config edit)", name)
+	case req.GPUs < 1 || req.GPUs > 64:
+		return GPUsChange{}, errors.New("a replica uses 1 to 64 cards")
+	}
+	i, err := gpuProfile(m, req.Profile)
+	if err != nil {
+		return GPUsChange{}, err
+	}
+	p := m.Profiles[i]
+	before := max(p.GPUsPerNode, p.TP, 1) * max(p.Nodes, p.PP, 1)
+	out := GPUsChange{Model: name, Profile: p.Name, Engine: p.Engine, Class: p.GPUClass, Before: before, After: req.GPUs}
+	if before == req.GPUs {
+		return out, fmt.Errorf("%s's %s profile already uses %d card(s)", name, p.Name, before)
+	}
+
+	sh, note := s.shapeOf(ctx, m)
+	next, err := respread(p, req.GPUs, sh)
+	if err != nil {
+		return out, err
+	}
+	out.Args = next.Args
+	switch kv := int64(contextOf(m, p)) * sh.kvPerToken; {
+	case sh.weights == 0 || sh.kvPerToken == 0 || contextOf(m, p) == 0:
+		if note == "" {
+			note = "its context is not known, so what each card holds is not checked"
+		}
+	default:
+		need := modelplan.CardNeed(p.Engine, sh.weights, kv, req.GPUs)
+		out.PerCard = fmt.Sprintf("%.1f GB a card of %d", need, req.GPUs)
+		if budget := modelplan.CardBudget(p.GPUClass); budget > 0 {
+			out.PerCard += fmt.Sprintf(", of %.0f usable", budget)
+			if need > budget {
+				return out, fmt.Errorf("on %d %s cards each would hold %.1f GB, more than the %.0f GB they hold: "+
+					"more cards, or a shorter context (llmgw ctl model context)", req.GPUs, p.GPUClass, need, budget)
+			}
+		}
+	}
+	out.Note = note
+
+	ids := s.replicasOf(name)
+	inPlace, err := s.roomFor(p.GPUClass, req.GPUs, name, len(ids) > 0, req.InPlace)
+	if err != nil {
+		return out, err
+	}
+	profiles := slices.Clone(m.Profiles)
+	profiles[i] = next
+	if _, err := s.deps.Models.SetProfiles(ctx, name, profiles, by); err != nil {
+		return out, err
+	}
+	if len(ids) == 0 {
+		return out, nil
+	}
+	restart := s.deps.Restart
+	if inPlace {
+		restart, out.InPlace = s.deps.RestartInPlace, true
+	}
+	if restart == nil {
+		return out, errors.New("the profile is changed, and this control plane cannot start its replicas again: llmgw ctl restart")
+	}
+	if err := restart(ids...); err != nil {
+		return out, fmt.Errorf("the profile is changed, and its replicas could not be started again on it: %w", err)
+	}
+	out.Restarted = len(ids)
+	return out, nil
+}
+
+// gpuProfile is the model's profile on cards to change: the one named, or
+// its only one.
+func gpuProfile(m catalog.Model, named string) (int, error) {
+	var onCards []int
+	for i, p := range m.Profiles {
+		if named != "" && p.Name == named {
+			if p.CPUOnly() {
+				return 0, fmt.Errorf("%s's %s profile runs on CPUs; give it a gpu_class first", m.Name, p.Name)
+			}
+			return i, nil
+		}
+		if !p.CPUOnly() {
+			onCards = append(onCards, i)
+		}
+	}
+	switch {
+	case named != "":
+		return 0, fmt.Errorf("%s has no profile %q", m.Name, named)
+	case len(onCards) == 0:
+		return 0, fmt.Errorf("%s runs on CPUs only; a profile on cards needs a gpu_class", m.Name)
+	case len(onCards) > 1:
+		var names []string
+		for _, i := range onCards {
+			names = append(names, m.Profiles[i].Name)
+		}
+		return 0, fmt.Errorf("%s has several profiles on cards (%s): name one", m.Name, strings.Join(names, ", "))
+	}
+	return onCards[0], nil
+}
+
+// respread is a profile spread over n cards of one node.
+func respread(p catalog.Profile, n int, sh shape) (catalog.Profile, error) {
+	for _, f := range tpFlags[p.Engine] {
+		p.Args = dropArg(p.Args, f)
+	}
+	switch p.Engine {
+	case "vllm", "sglang", "trtllm":
+		if !modelplan.SplitsHeads(sh.kvHeads, n) {
+			return p, fmt.Errorf("the model has %d KV heads, which tensor parallel cannot spread over %d cards: "+
+				"a number that divides them, or that they divide", sh.kvHeads, n)
+		}
+		p.TP, p.GPUsPerNode = n, n
+	case "llamacpp":
+		// llama.cpp has no tensor parallel: it splits the layers over a
+		// node's cards, in proportion to their free memory.
+		p.TP, p.GPUsPerNode = 1, n
+	default:
+		return p, fmt.Errorf("a %s replica runs on one card", p.Engine)
+	}
+	// One node: a replica spanning nodes is laid out by hand, with a fabric.
+	p.Nodes, p.PP = 1, 0
+	return p, nil
+}
+
+// contextOf is the tokens a profile's KV cache holds: llama.cpp's pool, or
+// the engine's limit, or the model's context.
+func contextOf(m catalog.Model, p catalog.Profile) int {
+	switch p.Engine {
+	case "llamacpp":
+		if n := argInt(p.Args, "--ctx-size"); n > 0 {
+			return n
+		}
+	case "vllm":
+		if n := argInt(p.Args, "--max-model-len"); n > 0 {
+			return n
+		}
+	case "sglang":
+		if n := argInt(p.Args, "--context-length"); n > 0 {
+			return n
+		}
+	}
+	return m.ContextLength
+}
+
+func (s *Server) replicasOf(model string) []string {
+	var ids []string
+	if s.deps.Fleet != nil {
+		for _, r := range s.deps.Fleet.Replicas() {
+			if r.Model == model {
+				ids = append(ids, r.ID)
+			}
+		}
+	}
+	return ids
+}
+
+// roomFor says whether the new replicas fit: a node with n cards of the
+// class at all, and free beside the running ones — or, asked to start them
+// in place, once the model's own are stopped. True is in place.
+func (s *Server) roomFor(class string, n int, model string, running, inPlace bool) (bool, error) {
+	if s.deps.Pool == nil {
+		return inPlace, nil
+	}
+	own := map[string]int{}
+	if s.deps.Fleet != nil {
+		for _, r := range s.deps.Fleet.Replicas() {
+			if r.Model == model {
+				for _, sh := range r.Shards {
+					own[sh.NodeID] += len(sh.GPUIndices)
+				}
+			}
+		}
+	}
+	most, freeMost, besideOK, inPlaceOK := 0, 0, false, false
+	for _, node := range s.deps.Pool() {
+		for _, k := range modelplan.Kinds(node) {
+			if k.Class != class {
+				continue
+			}
+			most, freeMost = max(most, k.GPUs), max(freeMost, k.FreeGPUs)
+			besideOK = besideOK || k.FreeGPUs >= n
+			inPlaceOK = inPlaceOK || k.FreeGPUs+own[node.ID] >= n
+		}
+	}
+	switch {
+	case most == 0:
+		return false, fmt.Errorf("no ready node has %s cards", class)
+	case most < n:
+		return false, fmt.Errorf("no node has %d %s cards; the most one has is %d", n, class, most)
+	case !running || besideOK:
+		return false, nil
+	case !inPlaceOK:
+		return false, fmt.Errorf("no node has %d %s cards free, even with this model's own stopped (%d at most free now): "+
+			"other models hold them", n, class, freeMost)
+	case !inPlace:
+		return false, fmt.Errorf("no node has %d %s cards free beside the running replica (%d at most); "+
+			"--in-place stops each replica first, and the model does not answer while the new one loads", n, class, freeMost)
+	}
+	return true, nil
+}
+
+// dropArg is arguments without a flag and its value.
+func dropArg(args []string, flag string) []string {
+	out := make([]string, 0, len(args))
+	for i := 0; i < len(args); i++ {
+		switch {
+		case args[i] == flag:
+			i++ // its value
+		case strings.HasPrefix(args[i], flag+"="):
+		default:
+			out = append(out, args[i])
+		}
+	}
+	return out
+}

internal/control/modelserve/gpus_test.go

@@ -0,0 +1,151 @@
+package modelserve
+
+import (
+	"context"
+	"slices"
+	"strings"
+	"testing"
+
+	"gitea.mynodes.xyz/ilker/llmgw/internal/control/catalog"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/control/reconciler"
+	"gitea.mynodes.xyz/ilker/llmgw/internal/modelplan"
+)
+
+// gpuSetup is the update setup with qwen on one L40S card of llama.cpp, and
+// a pool of one node of four, of which free are free.
+func gpuSetup(t *testing.T, free int) (*Server, *updModels, *restartFleet, *[]string) {
+	t.Helper()
+	s, _, models, _, fleet := updSetup(t)
+	m := models.models["qwen"]
+	m.ContextLength = 8192
+	m.Origin = &catalog.Origin{Repo: "o/r", Files: []catalog.OriginFile{{Path: "q4.gguf", Bytes: 20 << 30}}}
+	m.Profiles = []catalog.Profile{
+		{Name: "cpu", Engine: "llamacpp", MinMemoryGB: 8},
+		{Name: "gpu", Engine: "llamacpp", GPUClass: "l40s-48g", TP: 1, GPUsPerNode: 1,
+			Args: []string{"--ctx-size", "8192", "--tensor-split", "1"}},
+	}
+	models.models["qwen"] = m
+	s.deps.Pool = func() []modelplan.PoolNode {
+		return []modelplan.PoolNode{{ID: "g1", Class: "l40s-48g", GPUMemGB: 48, GPUs: 4, FreeGPUs: free}}
+	}
+	var inPlace []string
+	s.deps.RestartInPlace = func(ids ...string) error { inPlace = append(inPlace, ids...); return nil }
+	return s, models, fleet, &inPlace
+}
+
+func TestAModelIsSpreadOverMoreCardsAndItsReplicasStartedAgainBesideTheOld(t *testing.T) {
+	s, models, fleet, inPlace := gpuSetup(t, 3)
+	got, err := s.SetGPUs(context.Background(), "qwen", GPUsRequest{GPUs: 2}, "admin")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got.Profile != "gpu" || got.Before != 1 || got.After != 2 || got.Restarted != 2 || got.InPlace {
+		t.Fatalf("%+v", got)
+	}
+	p := models.models["qwen"].Profiles[1]
+	// llama.cpp splits layers over a node's cards: gpus_per_node, not tp,
+	// and the old card's proportions are gone.
+	if p.GPUsPerNode != 2 || p.TP != 1 || p.Nodes != 1 || !slices.Equal(p.Args, []string{"--ctx-size", "8192"}) {
+		t.Errorf("profile %+v", p)
+	}
+	if !strings.Contains(got.PerCard, "a card of 2") {
+		t.Errorf("per card %q", got.PerCard)
+	}
+	if len(*inPlace) != 0 {
+		t.Errorf("stopped first though there was room: %v", *inPlace)
+	}
+	for _, r := range fleet.Replicas() {
+		if !strings.Contains(r.ID, "-new") {
+			t.Errorf("%s was not started again", r.ID)
+		}
+	}
+	if models.models["qwen"].Profiles[0].GPUsPerNode != 0 {
+		t.Error("the CPU profile was changed too")
+	}
+}
+
+func TestWithNoRoomBesideIt_InPlaceIsAskedFor(t *testing.T) {
+	// One card free and the model's replicas on the others: two do not fit
+	// beside them.
+	s, _, fleet, inPlace := gpuSetup(t, 1)
+	fleet.reps[0].Shards = []reconciler.Shard{{NodeID: "g1", GPUIndices: []int{1}}}
+	fleet.reps[1].Shards = []reconciler.Shard{{NodeID: "g1", GPUIndices: []int{2}}}
+	_, err := s.SetGPUs(context.Background(), "qwen", GPUsRequest{GPUs: 2}, "admin")
+	if err == nil || !strings.Contains(err.Error(), "--in-place") {
+		t.Fatalf("not asked for in place: %v", err)
+	}
+	got, err := s.SetGPUs(context.Background(), "qwen", GPUsRequest{GPUs: 2, InPlace: true}, "admin")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !got.InPlace || len(*inPlace) != 2 {
+		t.Errorf("%+v, stopped first: %v", got, *inPlace)
+	}
+}
+
+func TestMoreCardsThanANodeHasOrOthersHoldAreRefused(t *testing.T) {
+	s, models, _, _ := gpuSetup(t, 4)
+	if _, err := s.SetGPUs(context.Background(), "qwen", GPUsRequest{GPUs: 8}, "admin"); err == nil ||
+		!strings.Contains(err.Error(), "the most one has is 4") {
+		t.Errorf("eight cards on a node of four: %v", err)
+	}
+	s, _, _, _ = gpuSetup(t, 0)
+	if _, err := s.SetGPUs(context.Background(), "qwen", GPUsRequest{GPUs: 2, InPlace: true}, "admin"); err == nil ||
+		!strings.Contains(err.Error(), "other models hold them") {
+		t.Errorf("cards other models hold: %v", err)
+	}
+	if models.models["qwen"].Profiles[1].GPUsPerNode != 1 {
+		t.Error("a refused change was kept")
+	}
+}
+
+func TestFewerCardsThanTheModelNeedsAreRefusedWithTheFigures(t *testing.T) {
+	s, models, _, _ := gpuSetup(t, 3)
+	m := models.models["qwen"]
+	m.Profiles[1].GPUsPerNode, m.Origin.Files[0].Bytes = 4, 120<<30
+	models.models["qwen"] = m
+	_, err := s.SetGPUs(context.Background(), "qwen", GPUsRequest{GPUs: 2}, "admin")
+	if err == nil || !strings.Contains(err.Error(), "GB") {
+		t.Errorf("120 GB on two 48 GB cards: %v", err)
+	}
+}
+
+func TestTensorParallelFollowsTheKVHeads(t *testing.T) {
+	p := catalog.Profile{Name: "gpu", Engine: "vllm", GPUClass: "h100-80g", TP: 2, GPUsPerNode: 2,
+		Args: []string{"--tensor-parallel-size", "2", "--max-model-len", "32768"}}
+	if _, err := respread(p, 3, shape{kvHeads: 8}); err == nil || !strings.Contains(err.Error(), "8 KV heads") {
+		t.Errorf("three ways for eight heads: %v", err)
+	}
+	next, err := respread(p, 4, shape{kvHeads: 8})
+	if err != nil {
+		t.Fatal(err)
+	}
+	// The agent sets the width from tp; the profile's own spelling went.
+	if next.TP != 4 || next.GPUsPerNode != 4 || !slices.Equal(next.Args, []string{"--max-model-len", "32768"}) {
+		t.Errorf("%+v", next)
+	}
+	for _, eng := range []string{"comfyui", "strata"} {
+		if _, err := respread(catalog.Profile{Name: "x", Engine: eng}, 2, shape{}); err == nil {
+			t.Errorf("%s spread over two cards", eng)
+		}
+	}
+}
+
+func TestAModelOnCPUsOrWithSeveralCardProfilesSaysWhy(t *testing.T) {
+	s, models, _, _ := gpuSetup(t, 3)
+	m := models.models["qwen"]
+	m.Profiles = append(m.Profiles, catalog.Profile{Name: "gpu2", Engine: "vllm", GPUClass: "l40s-48g", TP: 1})
+	models.models["qwen"] = m
+	if _, err := s.SetGPUs(context.Background(), "qwen", GPUsRequest{GPUs: 2}, "admin"); err == nil ||
+		!strings.Contains(err.Error(), "name one") {
+		t.Errorf("two profiles on cards and none named: %v", err)
+	}
+	if _, err := s.SetGPUs(context.Background(), "qwen", GPUsRequest{GPUs: 2, Profile: "cpu"}, "admin"); err == nil ||
+		!strings.Contains(err.Error(), "gpu_class") {
+		t.Errorf("a CPU profile: %v", err)
+	}
+	if _, err := s.SetGPUs(context.Background(), "qwen", GPUsRequest{GPUs: 1, Profile: "gpu"}, "admin"); err == nil ||
+		!strings.Contains(err.Error(), "already") {
+		t.Errorf("the same count: %v", err)
+	}
+}

internal/control/modelserve/modelserve.go

@@ -82,6 +82,9 @@ type Deps struct {
 	Current func() []catalog.Model
 	Models  Models
 	Restart func(ids ...string) error
+	// RestartInPlace stops each replica before its replacement starts, for
+	// when there is no room beside it.
+	RestartInPlace func(ids ...string) error
 
 	// Pool is the fleet's ready nodes as a plan sees them: what decides
 	// which file suits the hardware, and how it is spread.

internal/control/modelserve/updates.go

@@ -34,6 +34,8 @@ type Models interface {
 	Declared(name string) bool
 	SetSource(ctx context.Context, name, source string, origin *catalog.Origin, by string) (catalog.Model, error)
 	SetOrigin(ctx context.Context, name string, origin *catalog.Origin) error
+	SetContext(ctx context.Context, name string, contextLength, maxOutput int, profiles []catalog.Profile, by string) (catalog.Model, error)
+	SetProfiles(ctx context.Context, name string, profiles []catalog.Profile, by string) (catalog.Model, error)
 	// For a canary: the candidate added beside the model and taken out
 	// again, and the share of requests it is given.
 	Add(ctx context.Context, m catalog.Model, by string) error

internal/control/modelserve/updates_test.go

@@ -104,6 +104,23 @@ func (c *updModels) SetCanary(_ context.Context, n string, cn *catalog.Canary, _
 	c.models[n] = m
 	return m, nil
 }
+func (c *updModels) SetContext(_ context.Context, n string, ctxLen, maxOut int, ps []catalog.Profile, _ string) (catalog.Model, error) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	m := c.models[n]
+	m.ContextLength, m.MaxOutput, m.Profiles = ctxLen, maxOut, ps
+	c.models[n] = m
+	return m, nil
+}
+
+func (c *updModels) SetProfiles(_ context.Context, n string, ps []catalog.Profile, _ string) (catalog.Model, error) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	m := c.models[n]
+	m.Profiles = ps
+	c.models[n] = m
+	return m, nil
+}
 func (c *updModels) all() []catalog.Model {
 	c.mu.Lock()
 	defer c.mu.Unlock()

internal/control/modelsvc/modelsvc.go

@@ -154,6 +154,9 @@ func (s *Service) add(ctx context.Context, m catalog.Model, by string, onlyNew b
 		next = append(next, e)
 	}
 	next = append(next, m)
+	if err := m.CheckPickle(); err != nil {
+		return err
+	}
 
 	built, err := catalog.New(next)
 	if err != nil {
@@ -266,6 +269,17 @@ func (s *Service) Resources(ctx context.Context, name, profile string, memoryGB,
 // SetProfiles replaces a model's profiles, as a model added with them
 // would have: for a model whose engines are to run otherwise, its args among
 // it. Running replicas keep what they were started with until restarted.
+// SetContext changes how much one request may use and what it may answer,
+// with the profiles that carry it: their engines' arguments and memory.
+func (s *Service) SetContext(ctx context.Context, name string, contextLength, maxOutput int,
+	profiles []catalog.Profile, by string,
+) (catalog.Model, error) {
+	return s.update(ctx, name, by, "change its context", func(m *catalog.Model) error {
+		m.ContextLength, m.MaxOutput, m.Profiles = contextLength, maxOutput, profiles
+		return nil
+	})
+}
+
 func (s *Service) SetProfiles(ctx context.Context, name string, profiles []catalog.Profile, by string) (catalog.Model, error) {
 	if len(profiles) == 0 {
 		return catalog.Model{}, errors.New("a model needs a profile")
@@ -406,9 +420,13 @@ func (s *Service) update(ctx context.Context, name, by, what string, change func
 	found := false
 	for _, e := range stored {
 		if e.Name == name {
+			before := append(e.PickleSources(), e.UnlookedSources()...)
 			if err := change(&e); err != nil {
 				return catalog.Model{}, err
 			}
+			if err := e.CheckPickle(before...); err != nil {
+				return catalog.Model{}, err
+			}
 			found = true
 		}
 		next = append(next, e)

internal/docgen/reference_gen.go

@@ -230,6 +230,7 @@ var Reference = []Block{
 	}},
 	{Package: "config", Name: "Mirror", Title: "Mirror", Doc: "", Fields: []Field{
 		{Key: "api", Type: "string", Doc: "API is the mirror's management address, scheme included, for example\nhttps://models.internal:8082. Empty means no mirror is configured.\n\nThe management port, not the one nodes fetch bytes from. Those are two\nlisteners on the mirror precisely so this one can be locked down without\nre-certifying every agent's fetcher.", Nested: ""},
+		{Key: "peers", Type: "[]string", Doc: "Peers are the management addresses of mirrors kept in step with this\none (llmgw mirror serve --peer): asked when it cannot be reached, and\na node given the same file on one of them while it is down.", Nested: ""},
 		{Key: "tls", Type: "tlsconf.Config", Doc: "TLS authenticates to the mirror's management API, with the same CA the\nagents use. Empty leaves the connection in plain HTTP, which is only\nacceptable on a network where nothing untrusted can reach that port —\nthe mirror says as much in its own log when it starts that way.", Nested: "tlsconf.Config"},
 		{Key: "registry", Type: "string", Doc: "Registry is the model registry the mirror page browses, empty for the\npublic Hugging Face. Set it to an internal mirror of the registry where\nthe control plane cannot reach the internet — or leave the whole mirror\nblock out, which turns the explorer off rather than giving it a search\nbox that can only time out.", Nested: ""},
 		{Key: "timeout", Type: "time.Duration", Doc: "Timeout bounds a management call. It does not bound a download: the\nmirror answers a fetch as soon as it has started one, and the bytes keep\narriving long after this deadline has passed.", Nested: ""},
@@ -240,6 +241,7 @@ var Reference = []Block{
 		{Key: "aliases", Type: "[]string", Doc: "", Nested: ""},
 		{Key: "description", Type: "string", Doc: "Description is what this model is for, in a line.\n\nWritten by whoever added it, because nobody else can say it. A registry\noffers a pipeline tag and a wall of README, and neither answers the\nquestion a person picking from a list is asking: which of these nine\nnames should I send this request to.\n\nIt reaches clients through /v1/models, which is the only view most of\nthem ever see. A gateway serving nine models by name alone makes that\nchoice by guesswork.", Nested: ""},
 		{Key: "source", Type: "string", Doc: "Source is the weights path or a HuggingFace id.", Nested: ""},
+		{Key: "allow_pickle", Type: "bool", Doc: "AllowPickle lets the nodes fetch this model's weights, or its\nadapters', in a pickle format (.bin, .pt, .pth, .ckpt), which run code\nas they are loaded. Off, such a source is refused when the model is\nadded and by the node before it fetches anything. For a model that has\nno safetensors or GGUF form and whose files were checked.", Nested: ""},
 		{Key: "profiles", Type: "[]Profile", Doc: "", Nested: ""},
 		{Key: "scaling", Type: "Scaling", Doc: "", Nested: "catalog.Scaling"},
 		{Key: "class", Type: "string", Doc: "Class is how much this model's capacity is worth against another's,\nwhen there is not enough for both: critical, standard or opportunistic.\nEmpty is standard, or opportunistic for a model loaded on demand. See\nEffectiveClass.", Nested: ""},
@@ -503,6 +505,11 @@ var Reference = []Block{
 	{Package: "engines", Name: "SimConfig", Title: "SimConfig", Doc: "", Fields: []Field{
 		{Key: "bin", Type: "string", Doc: "Bin is the llmgw binary itself; the engine is one of its subcommands.\nEmpty means whatever started this agent, which is right almost always.", Nested: ""},
 	}},
+	{Package: "engines", Name: "StrataConfig", Title: "StrataConfig", Doc: "", Fields: []Field{
+		{Key: "dir", Type: "string", Doc: "Dir is Strata's source directory, where setup.py and serve/ are.", Nested: ""},
+		{Key: "python", Type: "string", Doc: "Python is the interpreter of its environment; Dir's .venv when unset.", Nested: ""},
+		{Key: "engine", Type: "string", Doc: "Engine is the directory of its prebuilt engine, where the strata\nbinary and the CUDA libraries it ships are; Dir's engine/ when unset.", Nested: ""},
+	}},
 	{Package: "engines", Name: "TRTLLMConfig", Title: "TRTLLMConfig", Doc: "", Fields: []Field{
 		{Key: "python", Type: "string", Doc: "Python is the interpreter of the environment TensorRT-LLM is installed\nin: <venv>/bin/python for a virtual environment, python3 in NVIDIA's\ncontainer. Naming it is what makes this node offer the engine; empty\noffers nothing.\n\nThe interpreter rather than trtllm-serve, because llmgw starts the\nserver through a few lines of its own that make it demand the\ngateway's key, which trtllm-serve has no option for.", Nested: ""},
 	}},
@@ -612,6 +619,7 @@ var Reference = []Block{
 		{Key: "trtllm", Type: "TRTLLMConfig", Doc: "TRTLLM is TensorRT-LLM, which the agent does not install: a node\noffers it only when its interpreter is named here.", Nested: "engines.TRTLLMConfig"},
 		{Key: "sim", Type: "SimConfig", Doc: "", Nested: "engines.SimConfig"},
 		{Key: "comfyui", Type: "ComfyUIConfig", Doc: "ComfyUI makes pictures; offered by a node whose ComfyUI is named here.", Nested: "engines.ComfyUIConfig"},
+		{Key: "strata", Type: "StrataConfig", Doc: "Strata serves a few mixture-of-experts models from RAM and one card;\noffered by a node whose Strata is named here, or that installed it.", Nested: "engines.StrataConfig"},
 	}},
 	{Package: "guard", Name: "Config", Title: "guard.Config", Doc: "", Fields: []Field{
 		{Key: "enabled", Type: "*bool", Doc: "Enabled turns every policy on or off at once: false checks nothing,\nwhatever is defined or attached. Unset is on.", Nested: ""},

internal/docgen/render.go

@@ -23,6 +23,7 @@ var Sources = []string{
 	"internal/agent/engines/trtllm.go",
 	"internal/agent/engines/sim.go",
 	"internal/agent/engines/comfyui.go",
+	"internal/agent/engines/strata.go",
 	"internal/agent/supervisor/supervisor.go",
 	"internal/tlsconf/tlsconf.go",
 	"internal/control/catalog/catalog.go",

internal/doctor/doctor.go

@@ -385,6 +385,12 @@ func Evaluate(f Facts) []Finding {
 				add(Warn, "gpu", fmt.Sprintf("GPU %s is kept out of placement: %s",
 					discovery.DeviceLabel(g.Index), strings.Join(g.Health.Problems, "; ")), "")
 			}
+			if g.MIG == nil {
+				if p := g.PCIe.Problem(); p != "" {
+					add(Warn, "gpu", fmt.Sprintf("GPU %d (%s) %s", g.Index, g.Name, p),
+						"lspci -vv -s "+g.PCIBusID+" shows LnkCap (the card) against LnkSta (the link)")
+				}
+			}
 		}
 	}
 	evaluateMIG(f, add)

internal/gguf/gguf.go

@@ -41,6 +41,9 @@ type Header struct {
 	// them; otherwise a head is EmbeddingLength / HeadCount.
 	KeyLength   uint64
 	ValueLength uint64
+	// ContextLength is the context the model was trained for
+	// (<arch>.context_length); 0 when the file does not say.
+	ContextLength uint64
 	// Embedding is whether the model pools its tokens into one vector
 	// (<arch>.pooling_type set and not none): an embedding model, which
 	// llama.cpp serves with --embedding and does not chat.
@@ -271,6 +274,8 @@ func Read(r io.Reader) (Header, error) {
 				h.ValueLength = num
 			case "pooling_type":
 				h.Embedding = num > 0
+			case "context_length":
+				h.ContextLength = num
 			}
 		}
 		// The tokenizer's arrays — megabytes of vocabulary — come after the

internal/guard/gate.go

@@ -418,7 +418,15 @@ func (g *Gate) Input(ctx context.Context, s Subject, body []byte) (Verdict, erro
 	// Outside services next, then the judge: what a pattern already stopped
 	// need not wait for either.
 	if user != "" {
-		sf, serr := g.askServices(ctx, p, true, user)
+		// As the model will read it: what redact took out stays out of the
+		// outside service's sight too.
+		told := user
+		for _, rw := range rewrites {
+			if rw.seg.text != "" && strings.Contains(told, rw.seg.text) {
+				told = strings.ReplaceAll(told, rw.seg.text, rw.text)
+			}
+		}
+		sf, serr := g.askServices(ctx, p, true, told)
 		if serr != nil && !v.Shadow {
 			v.Took = g.now().Sub(start)
 			return v, serr

internal/guard/guard.go

@@ -54,6 +54,10 @@ type Subject struct {
 	Stream bool
 	// Tools and Vision are what the model's catalog entry says it can do.
 	Tools, Vision bool
+	// Restored are the values the request's redaction took out, by
+	// placeholder, which restore puts back in the answer: outside services
+	// reading the answer are not told them.
+	Restored map[string]string
 }
 
 // Verdict is a judgement on one stage of one request.

internal/guard/output.go

@@ -68,27 +68,32 @@ func (g *Gate) Output(ctx context.Context, s Subject) OutputFilter {
 		return nil
 	}
 	memo := g.tools.get(s.RequestID)
-	return &outFilter{ctx: ctx, g: g, p: p, names: names, shadow: p.Mode == "shadow",
+	return &outFilter{ctx: ctx, g: g, p: p, names: names, shadow: p.Mode == "shadow", restored: s.Restored,
 		schemas: memo.schemas, answer: memo.answer, user: memo.user, choices: map[int]*choiceState{}}
 }
 
 // outFilter judges one answer.
 type outFilter struct {
-	ctx     context.Context
-	g       *Gate
-	p       Policy
-	names   string
-	shadow  bool
-	schemas map[string]json.RawMessage
-	answer  json.RawMessage
-	full    strings.Builder // the whole answer, when it must fit a schema
-	user    string          // the request's last user turn, for the judge
-	sofar   strings.Builder // the answer so far, for the judge
-	served  strings.Builder // the answer so far, for outside services, asked once at its end
-	judged  int             // how much of it the judge has read
-	v       Verdict
-	took    time.Duration
-	stopped bool
+	// servicesAsked says the outside services have read the answer.
+	servicesAsked bool
+	// restored are the request's redacted values, by placeholder, put back
+	// in the answer before this filter reads it.
+	restored map[string]string
+	ctx      context.Context
+	g        *Gate
+	p        Policy
+	names    string
+	shadow   bool
+	schemas  map[string]json.RawMessage
+	answer   json.RawMessage
+	full     strings.Builder // the whole answer, when it must fit a schema
+	user     string          // the request's last user turn, for the judge
+	sofar    strings.Builder // the answer so far, for the judge
+	served   strings.Builder // the answer so far, for outside services, asked once at its end
+	judged   int             // how much of it the judge has read
+	v        Verdict
+	took     time.Duration
+	stopped  bool
 
 	held     []string // events not yet released
 	sent     bool     // whether a chunk this filter made has gone out
@@ -103,6 +108,8 @@ type choiceState struct {
 	// replaced: it reaches the client only in a chunk this filter makes.
 	carried bool
 	calls   map[int]*call
+	// done says the choice has finished: its text is all here.
+	done bool
 }
 
 type call struct {
@@ -212,6 +219,13 @@ func (f *outFilter) askServices(text string) finding {
 	if !slices.ContainsFunc(f.p.Services, Service.onOutput) || text == "" {
 		return finding{}
 	}
+	// As the model wrote it: with restore, the values put back for the
+	// client are taken out again, as redact took them out of the request.
+	for ph, v := range f.restored {
+		if v != "" {
+			text = strings.ReplaceAll(text, v, ph)
+		}
+	}
 	start := time.Now()
 	fs, err := f.g.askServices(f.ctx, f.p, false, text)
 	f.took += time.Since(start)
@@ -379,6 +393,7 @@ func (f *outFilter) OnEvent(data string) ([]string, bool) {
 		}
 		if ch.FinishReason != nil && *ch.FinishReason != "" {
 			finished = true
+			st.done = true
 		}
 	}
 	if f.shadow {
@@ -445,8 +460,14 @@ func (f *outFilter) flush(final bool) (bool, map[int]string, bool) {
 		}
 		if i == 0 {
 			f.served.WriteString(head)
-			if final && f.askServices(f.served.String()).action == Block {
-				blocked = true
+			// Once, when the first choice is whole — not when another
+			// choice finishes first — and not again at the end of the
+			// stream after its finish.
+			if final && st.done && !f.servicesAsked {
+				f.servicesAsked = true
+				if f.askServices(f.served.String()).action == Block {
+					blocked = true
+				}
 			}
 		}
 		if head != "" {
@@ -566,6 +587,10 @@ func withoutContent(data string) string {
 }
 
 func (f *outFilter) OnEnd() []string {
+	// The stream is over: whatever a choice has is all it will have.
+	for _, st := range f.choices {
+		st.done = true
+	}
 	switch {
 	case f.stopped:
 		return nil

internal/guard/policy.go

@@ -220,10 +220,21 @@ func Merge(names []string, ps []Policy) Policy {
 		out.RequireCapabilities = out.RequireCapabilities || p.RequireCapabilities
 		out.ValidateStructured = out.ValidateStructured || p.ValidateStructured
 		out.Judge = mergeJudge(out.Judge, p.Judge)
-		// Every service any policy asks, each once.
+		// Every service any policy asks, each once; one named by two is
+		// asked the stricter way of the two: both directions either asks,
+		// blocking where either blocks.
 		for _, sv := range p.Services {
-			if !slices.ContainsFunc(out.Services, func(e Service) bool { return e.Name == sv.Name }) {
+			i := slices.IndexFunc(out.Services, func(e Service) bool { return e.Name == sv.Name })
+			if i < 0 {
 				out.Services = append(out.Services, sv)
+				continue
+			}
+			e := &out.Services[i]
+			if e.action() == Flag && sv.action() == Block {
+				e.Action = "block"
+			}
+			if e.onInput() != sv.onInput() || e.onOutput() != sv.onOutput() {
+				e.On = "both"
 			}
 		}
 		out.MaxOutputTokens = minSet(out.MaxOutputTokens, p.MaxOutputTokens)

internal/guard/service.go

@@ -144,7 +144,8 @@ func (s Service) ask(ctx context.Context, text string) (unsafe bool, reason stri
 		return false, "", fmt.Errorf("answer is not JSON: %w", err)
 	}
 	v, ok := at(doc, s.Blocked)
-	if !ok {
+	if !ok || v == nil {
+		// null is no answer: a service that cannot say is not saying safe.
 		return false, "", fmt.Errorf("the answer has no %s", s.Blocked)
 	}
 	if s.Reason != "" {

internal/guard/service_test.go

@@ -98,3 +98,89 @@ func TestAServiceIsCheckedWhenThePolicyIsRead(t *testing.T) {
 		t.Errorf("at: %v %v", v, ok)
 	}
 }
+
+// recorder is a service that finds nothing unsafe and keeps what it read.
+func recorder(t *testing.T) (*httptest.Server, *[]string) {
+	var seen []string
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		var in struct{ Input string }
+		_ = json.NewDecoder(r.Body).Decode(&in)
+		seen = append(seen, in.Input)
+		_, _ = w.Write([]byte(`{"flagged": false}`))
+	}))
+	t.Cleanup(srv.Close)
+	return srv, &seen
+}
+
+func TestAnOutsideServiceReadsWhatTheModelReadsNotWhatWasRedacted(t *testing.T) {
+	srv, seen := recorder(t)
+	g := gate(t, lib(map[string]Policy{"p": {
+		Input:    Rules{PII: map[string]string{"email": "redact"}},
+		Services: []Service{{Name: "cls", URL: srv.URL, Blocked: "flagged"}},
+	}}, "p"))
+	if _, err := g.Input(context.Background(), Subject{}, chat("user", "mail ayse@example.com")); err != nil {
+		t.Fatal(err)
+	}
+	if len(*seen) != 1 || strings.Contains((*seen)[0], "ayse@") || !strings.Contains((*seen)[0], "[EMAIL_1]") {
+		t.Errorf("the service read %q", *seen)
+	}
+}
+
+func TestAStreamedAnswerIsReadByAServiceOnce(t *testing.T) {
+	srv, seen := recorder(t)
+	g := gate(t, lib(map[string]Policy{"p": {
+		Services: []Service{{Name: "cls", URL: srv.URL, Blocked: "flagged", On: "output"}},
+	}}, "p"))
+	got, _, _ := stream(g.Output(context.Background(), Subject{Stream: true}), "Hello ", "there.")
+	if got != "Hello there." {
+		t.Errorf("client got %q", got)
+	}
+	if len(*seen) != 1 || (*seen)[0] != "Hello there." {
+		t.Errorf("the service was asked %d times: %q", len(*seen), *seen)
+	}
+}
+
+func TestAServiceNamedByTwoPoliciesIsAskedTheStricterWay(t *testing.T) {
+	a := Policy{Services: []Service{{Name: "cls", URL: "https://c", Blocked: "x", On: "input", Action: "flag"}}}
+	b := Policy{Services: []Service{{Name: "cls", URL: "https://c", Blocked: "x", On: "output", Action: "block"}}}
+	got := Merge([]string{"a", "b"}, []Policy{a, b}).Services
+	if len(got) != 1 || got[0].action() != Block || !got[0].onInput() || !got[0].onOutput() {
+		t.Errorf("%+v", got)
+	}
+}
+
+func TestAnOutputServiceReadsTheAnswerWithoutTheRestoredValues(t *testing.T) {
+	srv, seen := recorder(t)
+	g := gate(t, lib(map[string]Policy{"p": {
+		Services: []Service{{Name: "cls", URL: srv.URL, Blocked: "flagged", On: "output"}},
+	}}, "p"))
+	f := g.Output(context.Background(), Subject{Restored: map[string]string{"[EMAIL_1]": "ayse@example.com"}})
+	f.Buffered([]byte(`{"choices":[{"index":0,"message":{"role":"assistant","content":"Sent to ayse@example.com."},"finish_reason":"stop"}]}`))
+	if len(*seen) != 1 || strings.Contains((*seen)[0], "ayse@") || !strings.Contains((*seen)[0], "[EMAIL_1]") {
+		t.Errorf("the service read %q", *seen)
+	}
+}
+
+func TestOfTwoChoicesTheServiceReadsTheFirstWhole(t *testing.T) {
+	// The second finishes first; the first is still arriving then.
+	srv, seen := recorder(t)
+	g := gate(t, lib(map[string]Policy{"p": {
+		Services: []Service{{Name: "cls", URL: srv.URL, Blocked: "flagged", On: "output"}},
+	}}, "p"))
+	f := g.Output(context.Background(), Subject{Stream: true})
+	ev := func(idx int, text string, finish bool) string {
+		c := map[string]any{"index": idx, "delta": map[string]any{"content": text}, "finish_reason": nil}
+		if finish {
+			c["finish_reason"] = "stop"
+		}
+		b, _ := json.Marshal(map[string]any{"choices": []any{c}})
+		return string(b)
+	}
+	for _, e := range []string{ev(0, "The first ", false), ev(1, "Second.", true), ev(0, "answer.", true)} {
+		f.OnEvent(e)
+	}
+	f.OnEnd()
+	if len(*seen) != 1 || (*seen)[0] != "The first answer." {
+		t.Errorf("the service read %q", *seen)
+	}
+}

internal/hf/hf.go

@@ -18,6 +18,7 @@ import (
 	"net/http"
 	"net/url"
 	"regexp"
+	"slices"
 	"strings"
 	"sync"
 	"time"
@@ -69,15 +70,78 @@ type Model struct {
 	// for one that has none.
 	GGUF *GGUFInfo `json:"gguf"`
 	// Safetensors is the registry's summary of original weights.
-	Safetensors *struct {
-		Total int64 `json:"total"`
-	} `json:"safetensors,omitempty"`
+	Safetensors *SafetensorsInfo `json:"safetensors,omitempty"`
 	// Siblings are the repository's files, when asked for with WithFiles.
 	Siblings []struct {
 		RFilename string `json:"rfilename"`
 	} `json:"siblings,omitempty"`
 }
 
+// SafetensorsInfo is the registry's summary of a repository's safetensors.
+type SafetensorsInfo struct {
+	Total int64 `json:"total"`
+	// Parameters counts them by dtype: {"BF16": 8e9}. A packed quantisation
+	// (AWQ, GPTQ, MLX) is counted as its I32 or U32 at the count of weights
+	// it unpacks to, not of the integers that hold them.
+	Parameters map[string]int64 `json:"parameters,omitempty"`
+}
+
+// dtypeBytes is what one parameter of a safetensors dtype takes.
+var dtypeBytes = map[string]float64{
+	"F64": 8, "I64": 8, "U64": 8, "F32": 4, "BF16": 2, "F16": 2, "I16": 2, "U16": 2,
+	"F8_E4M3": 1, "F8_E5M2": 1, "F8_E8M0": 1, "I8": 1, "U8": 1, "BOOL": 1,
+}
+
+// WeightBytes is about how much the safetensors weights take: each dtype's
+// count at its size, and a packed quantisation at 4 bits a weight, 8 for
+// one tagged 8-bit — I32 or U32 always, U8 for MXFP4 and NVFP4, which the
+// registry counts at the weights they unpack to. 0 when the registry has
+// no summary.
+func (m Model) WeightBytes() int64 {
+	if m.Safetensors == nil {
+		return 0
+	}
+	if len(m.Safetensors.Parameters) == 0 {
+		return m.Safetensors.Total * 2
+	}
+	packed := 0.5
+	if m.HasTag("8-bit") {
+		packed = 1
+	}
+	fp4 := m.HasTag("mxfp4") || m.HasTag("nvfp4") || m.HasTag("fp4")
+	var b float64
+	for dt, n := range m.Safetensors.Parameters {
+		switch per, ok := dtypeBytes[dt]; {
+		case dt == "U8" && fp4:
+			b += float64(n) * 0.5
+		case ok:
+			b += float64(n) * per
+		case dt == "I32" || dt == "U32":
+			b += float64(n) * packed
+		default:
+			b += float64(n) * 2
+		}
+	}
+	return int64(b)
+}
+
+// FP8 reports whether the safetensors weights are in an 8-bit float, which
+// vLLM runs only on cards of compute capability 8.9 or newer.
+func (m Model) FP8() bool {
+	if m.Safetensors == nil {
+		return false
+	}
+	for dt := range m.Safetensors.Parameters {
+		if strings.HasPrefix(dt, "F8_") {
+			return true
+		}
+	}
+	return false
+}
+
+// HasTag reports whether the registry tags the repository so.
+func (m Model) HasTag(tag string) bool { return slices.Contains(m.Tags, tag) }
+
 // GGUFInfo is the registry's summary of a repository's GGUF weights.
 type GGUFInfo struct {
 	// Total is the parameter count.

internal/mirror/api.go

@@ -52,6 +52,9 @@ type ModelsResponse struct {
 	// when it is not.
 	Base   string   `json:"base"`
 	Models []Status `json:"models"`
+	// Removed are the versions removed here, model/version to when (ms),
+	// for a peer mirror to remove too.
+	Removed map[string]int64 `json:"removed,omitempty"`
 }
 
 func (a *API) list(w http.ResponseWriter, _ *http.Request) {
@@ -60,7 +63,7 @@ func (a *API) list(w http.ResponseWriter, _ *http.Request) {
 		fail(w, http.StatusInternalServerError, err)
 		return
 	}
-	writeJSON(w, http.StatusOK, ModelsResponse{Base: a.base, Models: all})
+	writeJSON(w, http.StatusOK, ModelsResponse{Base: a.base, Models: all, Removed: a.m.Tombstones()})
 }
 
 // FetchRequest asks the mirror to download a model.

internal/mirror/client.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"io"
 	"net/http"
@@ -24,6 +25,29 @@ import (
 type Client struct {
 	base string
 	hc   *http.Client
+	// others are the mirrors kept in step with this one, asked in turn when
+	// it cannot be reached: the same models, on another machine.
+	others []string
+}
+
+// NewFailoverClient is NewClient over several mirrors kept in step: the
+// first answers, and the next is asked when it cannot be reached.
+func NewFailoverClient(addrs []string, tc tlsconf.Config, timeout time.Duration) (*Client, error) {
+	if len(addrs) == 0 {
+		return nil, fmt.Errorf("no mirror address")
+	}
+	c, err := NewClient(addrs[0], tc, timeout)
+	if err != nil {
+		return nil, err
+	}
+	for _, a := range addrs[1:] {
+		u, err := url.Parse(a)
+		if err != nil || u.Host == "" {
+			return nil, fmt.Errorf("the mirror address %q is not a URL with a host", a)
+		}
+		c.others = append(c.others, strings.TrimSuffix(a, "/"))
+	}
+	return c, nil
 }
 
 // NewClient dials a mirror's management API, with mutual TLS when configured.
@@ -149,28 +173,42 @@ func (c *Client) Remove(ctx context.Context, model, version string) ([]string, e
 }
 
 func (c *Client) do(ctx context.Context, method, path string, body, out any) error {
-	var rdr io.Reader
+	var b []byte
 	if body != nil {
-		b, err := json.Marshal(body)
+		var err error
+		if b, err = json.Marshal(body); err != nil {
+			return err
+		}
+	}
+	// The mirror first, then those kept in step with it, while each cannot
+	// be reached at all: one that answers, even with a refusal, is the one.
+	var resp *http.Response
+	var errs []error
+	for _, base := range append([]string{c.base}, c.others...) {
+		var rdr io.Reader
+		if body != nil {
+			rdr = bytes.NewReader(b)
+		}
+		req, err := http.NewRequestWithContext(ctx, method, base+path, rdr)
 		if err != nil {
 			return err
 		}
-		rdr = bytes.NewReader(b)
+		if body != nil {
+			req.Header.Set("Content-Type", "application/json")
+		}
+		if resp, err = c.hc.Do(req); err == nil {
+			break
+		}
+		// Wrapped with the address because the failure an operator sees on
+		// a page is otherwise "connection refused" with no clue which of
+		// the machines refused it.
+		errs = append(errs, fmt.Errorf("reaching the mirror at %s: %w", base, err))
+		if ctx.Err() != nil {
+			break
+		}
 	}
-	req, err := http.NewRequestWithContext(ctx, method, c.base+path, rdr)
-	if err != nil {
-		return err
-	}
-	if body != nil {
-		req.Header.Set("Content-Type", "application/json")
-	}
-
-	resp, err := c.hc.Do(req)
-	if err != nil {
-		// Wrapped with the address because the failure an operator sees on a
-		// page is otherwise "connection refused" with no clue which of the two
-		// machines refused it.
-		return fmt.Errorf("reaching the mirror at %s: %w", c.base, err)
+	if resp == nil {
+		return errors.Join(errs...)
 	}
 	defer resp.Body.Close()
 

internal/mirror/manager.go

@@ -43,6 +43,9 @@ type Status struct {
 
 	StartedAt  time.Time `json:"started_at,omitzero"`
 	FinishedAt time.Time `json:"finished_at,omitzero"`
+	// AddedAt is when the file came to this mirror, as its modification
+	// time says: set against a removal elsewhere, the later stands.
+	AddedAt time.Time `json:"added_at,omitzero"`
 }
 
 // Percent reports progress, and false when the total is unknown.
@@ -106,6 +109,16 @@ var ErrAlreadyFetching = errors.New("a download for that version is already runn
 // the API call that asked for it, which is the point of doing it here rather
 // than making the caller hold a connection open for an hour.
 func (m *Manager) Start(source string, e Entry) (Status, error) {
+	return m.start(source, e, "", m.client)
+}
+
+// startCopy fetches a file a peer mirror holds, from a private address,
+// and keeps it only if its checksum is the peer's.
+func (m *Manager) startCopy(source string, e Entry, sha string) (Status, error) {
+	return m.start(source, e, sha, fetchClient(true))
+}
+
+func (m *Manager) start(source string, e Entry, expect string, client *http.Client) (Status, error) {
 	if source == "" {
 		return Status{}, fmt.Errorf("a source URL is required")
 	}
@@ -132,16 +145,16 @@ func (m *Manager) Start(source string, e Entry) (Status, error) {
 	out := *st
 	m.mu.Unlock()
 
-	go m.run(source, e, key)
+	go m.run(source, e, key, expect, client)
 	return out, nil
 }
 
-func (m *Manager) run(source string, e Entry, key string) {
+func (m *Manager) run(source string, e Entry, key, expect string, client *http.Client) {
 	// Its own context, deliberately unbounded by any request. A download is
 	// minutes to hours and the thing that asked for it is long gone.
 	ctx := context.Background()
 
-	added, err := addWith(ctx, m.client, m.root, source, e, func(done, total int64) {
+	added, err := addWith(ctx, client, m.root, source, e, func(done, total int64) {
 		m.mu.Lock()
 		if st, ok := m.running[key]; ok {
 			st.BytesDone, st.BytesTotal = done, total
@@ -149,6 +162,16 @@ func (m *Manager) run(source string, e Entry, key string) {
 		m.mu.Unlock()
 	})
 
+	if err == nil && expect != "" && added.SHA256 != expect {
+		// Not what the peer holds: a file changed or cut short on the way.
+		_ = Remove(m.root, e)
+		err = fmt.Errorf("the copy's sha256 %s is not the peer's %s", added.SHA256, expect)
+	}
+	if err == nil {
+		// Added again after it was removed: the new one stands.
+		m.clearTombstone(e.Model, e.Version)
+	}
+
 	m.mu.Lock()
 	defer m.mu.Unlock()
 	st, ok := m.running[key]
@@ -284,6 +307,9 @@ func (m *Manager) All() ([]Status, error) {
 			Model: e.Model, Version: e.Version, File: e.File,
 			State: Present, BytesDone: e.Bytes, BytesTotal: e.Bytes,
 		}
+		if fi, err := os.Stat(filepath.Join(m.root, e.Model, e.Version, e.File)); err == nil {
+			st.AddedAt = fi.ModTime()
+		}
 		// The checksum and the source are not recoverable from a directory
 		// listing, and the checksum is the one field worth having: it is what
 		// an operator compares against what upstream published. Dropping it in
@@ -364,6 +390,9 @@ func (m *Manager) RemoveVersion(model, version string) ([]string, error) {
 		return nil, fmt.Errorf("%s/%s: %w", model, version, os.ErrNotExist)
 	}
 	m.Forget(model, version)
+	// Kept, so a peer mirror that still has it removes it too, and does
+	// not copy it back here.
+	m.recordTombstone(model, version, time.Now())
 	m.log.Info("mirror removed a version", "model", model, "version", version, "files", len(removed))
 	return removed, nil
 }

internal/mirror/peer.go

@@ -0,0 +1,124 @@
+package mirror
+
+import (
+	"context"
+	"log/slog"
+	"strings"
+	"time"
+)
+
+// Mirrors kept in step, so the cluster has its models while one machine is
+// down: each reads its peers every while, copies the versions a peer has
+// that it lacks, checksum checked, and removes those a peer removed after
+// they came here. Neither leads; what was done last stands.
+
+// PeerInterval is how often each peer is read.
+const PeerInterval = 30 * time.Second
+
+// Peers keeps one mirror in step with others.
+type Peers struct {
+	M       *Manager
+	Clients []*Client
+	Log     *slog.Logger
+}
+
+// Run reads the peers until ctx ends.
+func (p *Peers) Run(ctx context.Context) {
+	t := time.NewTicker(PeerInterval)
+	defer t.Stop()
+	for {
+		for _, c := range p.Clients {
+			p.SyncOnce(ctx, c)
+		}
+		select {
+		case <-ctx.Done():
+			return
+		case <-t.C:
+		}
+	}
+}
+
+// SyncOnce brings this mirror in step with one peer.
+func (p *Peers) SyncOnce(ctx context.Context, c *Client) {
+	ctx, cancel := context.WithTimeout(ctx, 20*time.Second)
+	defer cancel()
+	theirs, err := c.List(ctx)
+	if err != nil {
+		p.Log.Warn("could not read a peer mirror; trying again shortly", "peer", c.Addr(), "err", err)
+		return
+	}
+	mine, err := p.M.All()
+	if err != nil {
+		p.Log.Warn("could not read this mirror", "err", err)
+		return
+	}
+	here := map[string]Status{}
+	for _, st := range mine {
+		here[st.Model+"/"+st.Version+"/"+st.File] = st
+	}
+	removed := p.M.Tombstones()
+
+	// What the peer removed after it came here goes here too.
+	for key, at := range theirs.Removed {
+		model, version, ok := strings.Cut(key, "/")
+		if !ok {
+			continue
+		}
+		newest := time.Time{}
+		has := false
+		for _, st := range mine {
+			if st.Model == model && st.Version == version {
+				has = true
+				if st.AddedAt.After(newest) {
+					newest = st.AddedAt
+				}
+			}
+		}
+		if has && newest.UnixMilli() < at {
+			if _, err := p.M.RemoveVersion(model, version); err == nil {
+				p.Log.Info("removed a version a peer mirror removed", "peer", c.Addr(), "model", model, "version", version)
+			}
+		} else if !has && at > removed[key] {
+			p.M.recordTombstone(model, version, time.UnixMilli(at))
+		}
+	}
+
+	// Versions whole there and missing here: every file present, each with
+	// its checksum, so a copy can be checked.
+	type version struct {
+		files []Status
+		whole bool
+	}
+	versions := map[string]*version{}
+	for _, st := range theirs.Models {
+		key := st.Model + "/" + st.Version
+		v := versions[key]
+		if v == nil {
+			v = &version{whole: true}
+			versions[key] = v
+		}
+		v.files = append(v.files, st)
+		v.whole = v.whole && st.State == Present && st.SHA256 != ""
+	}
+	for key, v := range versions {
+		if !v.whole {
+			continue
+		}
+		for _, st := range v.files {
+			if _, ok := here[key+"/"+st.File]; ok {
+				continue
+			}
+			// Removed here after the peer had it: not copied back. The
+			// peer removes its own when it reads this one.
+			if at, gone := removed[key]; gone && st.AddedAt.UnixMilli() < at {
+				continue
+			}
+			source := strings.TrimRight(theirs.Base, "/") + "/" + st.Model + "/" + st.Version + "/" + st.File
+			e := Entry{Model: st.Model, Version: st.Version, File: st.File}
+			if _, err := p.M.startCopy(source, e, st.SHA256); err == nil {
+				p.Log.Info("copying a version from a peer mirror", "peer", c.Addr(), "model", st.Model,
+					"version", st.Version, "file", st.File)
+			}
+		}
+	}
+}