Ilker ManapandClaude Opus 5.5
d35630defe
Collections owned by an organization, team, project or person; grants that add up down the tree and never cross an organization; chunks in PostgreSQL with pgvector, partitioned by organization, an HNSW index per collection, and row-level security as a second line behind each query's filter. Search fuses nearest-neighbour and full-text ranks. Off on SQLite and without pgvector, saying why. server init and the HA database setup install and enable it. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
56 lines
2.2 KiB
SQL
56 lines
2.2 KiB
SQL
-- Knowledge bases: collections of a company's documents that requests are
|
|
-- answered from (BACKLOG F). PostgreSQL only: there is no SQLite file of this
|
|
-- number, and on SQLite the feature is off.
|
|
--
|
|
-- The chunks themselves are not here. They need the pgvector extension, which
|
|
-- only a superuser can create and which a cluster may not have installed; a
|
|
-- migration that failed for want of it would stop the control plane from
|
|
-- starting at all. The store creates kb_chunks when it opens, if it can
|
|
-- (knowledge.go), and says why not when it cannot.
|
|
|
|
-- A collection has one owner in the tenancy tree and lives inside one
|
|
-- organization, which no grant can cross.
|
|
CREATE TABLE kb_collections (
|
|
id TEXT PRIMARY KEY,
|
|
org_id TEXT NOT NULL REFERENCES organizations(id),
|
|
owner_scope TEXT NOT NULL,
|
|
owner_id TEXT NOT NULL,
|
|
name TEXT NOT NULL,
|
|
-- Vectors from two models are not comparable, so a collection is embedded
|
|
-- by one model and its dimension is fixed with it.
|
|
embed_model TEXT NOT NULL,
|
|
dims INTEGER NOT NULL,
|
|
created_at BIGINT NOT NULL,
|
|
created_by TEXT NOT NULL DEFAULT '',
|
|
UNIQUE (org_id, name)
|
|
);
|
|
CREATE INDEX kb_collections_by_owner ON kb_collections(owner_scope, owner_id);
|
|
|
|
-- Who else may read or write a collection. Like guard_attachments, keyed by
|
|
-- scope with everything below the scope included; unlike guard policies,
|
|
-- grants add up rather than narrow.
|
|
CREATE TABLE kb_grants (
|
|
scope TEXT NOT NULL,
|
|
scope_id TEXT NOT NULL,
|
|
collection TEXT NOT NULL REFERENCES kb_collections(id) ON DELETE CASCADE,
|
|
perm TEXT NOT NULL,
|
|
added_at BIGINT NOT NULL,
|
|
added_by TEXT NOT NULL DEFAULT '',
|
|
PRIMARY KEY (scope, scope_id, collection)
|
|
);
|
|
CREATE INDEX kb_grants_by_collection ON kb_grants(collection);
|
|
|
|
-- A document as added, by name within its collection. The digest is what
|
|
-- decides whether adding it again changes anything.
|
|
CREATE TABLE kb_documents (
|
|
id TEXT PRIMARY KEY,
|
|
collection TEXT NOT NULL REFERENCES kb_collections(id) ON DELETE CASCADE,
|
|
name TEXT NOT NULL,
|
|
digest TEXT NOT NULL,
|
|
bytes BIGINT NOT NULL,
|
|
chunks INTEGER NOT NULL,
|
|
added_at BIGINT NOT NULL,
|
|
added_by TEXT NOT NULL DEFAULT '',
|
|
UNIQUE (collection, name)
|
|
);
|