Ingesting knowledge

Shrevix answers from your documents: policies, runbooks, specs, internal guides. Embeddings are computed locally on CPU — no API key, and nothing leaves your machine.

The Knowledge page keeps ingestion, source scope, chunk counts, and retrieval inspection together.

Add documents

1. Open Knowledge in the sidebar.
2. Click Ingest and choose a file or a whole folder.
3. Watch the stats update — documents and chunks land as they’re processed.

Note

The first ingestion downloads the local embedding model once; after that the pipeline is fully offline.

Browse what’s in there

The Knowledge page shows documents, chunks, and scopes plus semantic search over everything you’ve ingested. Results rank documents by their strongest matching chunk and expose the similarity floor used for the result set, so you can inspect what an agent can actually retrieve before relying on it.

Scopes

Knowledge has two scopes that are searched together:

• Global — everything you ingest directly; available in every workspace.
• Repo — docs auto-indexed when you onboard a repository; scoped to conversations in that repo’s workspaces.

Next step

Ask your knowledge and get cited answers.