Skip to content

Knowledge base

The company Knowledge Base is a curated store of documents that your agents can draw on while they work. It is a top-level, in-app feature: open a company and click Knowledge in the nav, or go directly to /companies/{id}/knowledge. From there you list, create, edit, and archive company-knowledge docs.

A doc is just titled markdown with a category. On its own it is searchable reference material for humans. Turn on the features below and the same docs become grounding context that agents receive automatically.

Semantic retrieval (RAG)

Set TRIMUS_EMBEDDING_ENABLED=true (worker and engine) to embed knowledge docs and retrieve them at run time.

  • A background worker embeds each doc's chunks into pgvector. Uploads and edits re-embed automatically; docs are re-scanned when their content changes.
  • When an agent runs an issue, the engine finds the most relevant chunks for that agent + issue and injects them into the prompt as a ## Company Knowledge section. The agent sees only the slice that matters for the work in front of it, not the whole library.

With the flag off, docs still list and edit normally — they just are not embedded or injected. The flag is runtime-editable from the operator control panel; see the TRIMUS_EMBEDDING_ENABLED row in the environment reference. The same flag also powers semantic search over notes.

Git-sync

Rather than paste docs by hand, you can import them from a repository. On the Knowledge page, configure a git source per company: a repo URL, an optional branch, an optional path glob, and (for private repos) an auth token. The token is encrypted at rest and never displayed back — the page only shows whether one is set.

  • Scheduled import. Set TRIMUS_KNOWLEDGE_GIT_SYNC_ENABLED=true (worker) and a background worker clones each company's configured repo on a schedule and imports matching files as company-knowledge docs.
  • Sync now. The manual Sync now button on the Knowledge page runs a single import in the background regardless of the flag, so you can pull the latest without waiting for the scheduler or enabling it at all.

See the TRIMUS_KNOWLEDGE_GIT_SYNC_ENABLED row in the environment reference.

The Knowledge Base is the company-tier hub of a small family of features that share the same embedding pipeline and the same ## Company Knowledge prompt injection.

  • Project-scoped knowledge. The same pipeline scoped to a single project's linked files, so an agent working a project issue gets that project's material instead of (or alongside) the company library. See projects and workspaces.
  • Slack ingestion. Admin-selected Slack channels can be ingested into the same company Knowledge Base — each thread becomes a knowledge doc, re-scanned periodically. See the Slack channel page.
  • Compiled Knowledge. A compile-time complement rather than a retrieval one: it distils company, project, and agent sources into typed, cross-linked wiki articles and injects a compact, always-on index into every prompt. RAG answers "what is relevant to this issue?"; Compiled Knowledge answers "what should every agent always know?" See compiled knowledge.

Operator checklist

  • Decide whether agents should be grounded on knowledge at all. If yes, set TRIMUS_EMBEDDING_ENABLED=true on both the worker and the engine.
  • Seed the Knowledge Base — write docs in-app, or configure a git source and click Sync now.
  • For hands-off imports, set TRIMUS_KNOWLEDGE_GIT_SYNC_ENABLED=true on the worker.
  • Consider the sibling features — project-scoped knowledge, Slack ingestion, and compiled knowledge — for coverage the company-tier docs do not give you.