Core concept
Knowledge sources and retrieval
Knowledge is stored in PostgreSQL, split into tenant-scoped chunks, indexed for full-text search, embedded, and retrieved through a hybrid lexical/vector path.
Articles and files
Publish articles or upload PDF, DOCX, TXT, Markdown, CSV, JSON, and HTML files up to 10 MB.
Website sources
Import a page or crawl up to 12 same-origin public pages with SSRF, DNS, redirect, size, and timeout controls.
Hybrid retrieval
Published chunks are ranked from full-text and vector candidates. Retrieval selects evidence; it does not make model output infallible.
Knowledge API
GET /api/v1/knowledge lists articles with cursor pagination and an optional state filter. POST /api/v1/knowledge creates a draft or published article and requires knowledge:write plus an Idempotency-Key. File and website imports remain dashboard workflows because they run as durable background jobs.