Vectors

Turns your files into a searchable index and answers questions with the nearest passages from it. It has no panel of its own, because the surfaces belong to the extensions built on top of it.

What you can do

• Get semantic search that finds a passage by meaning, not just by matching words.
• Have it back the Knowledge catalog, the assistant's project context, the Semantic toggle in search, and capability matching in the generation pipeline.
• Keep the whole thing on your machine by choosing an on-machine embedder, or send text to a remote one when you would rather.
• Rebuild an index incrementally as your files change, rather than from scratch.
• Get an answer even when the index is missing or the embedder stopped responding, because the query falls back to keyword search and says so.
• Check that retrieval works with Vectors: Run Retrieval Probe, which asks a question about one notebook and reports whether the right cell came back.

There is no vector database to run and no RAG framework here. It produces and serves an index over your own material.

Getting started

Choose your embedder first, in Settings under Vectors. Then turn on the Semantic toggle in search, or open the Knowledge catalog, and Vectors works through them. Vectors: Build Index, Vectors: Index Status and Vectors: Test Embedder are in the command palette when you want to work with an index directly.

Requirements

• For an embedder that runs on your machine: the desktop app, the Ollama Runtime extension, and an embedding model pulled once, such as nomic-embed-text.
• For a remote embedder: the address of an embedding endpoint, with its token kept in the SECRETS panel.

Good to know

• The default embedder matches shared words, not meaning. It proves the pipeline runs, not that retrieval works, so choose a real model for real results.
• Only the on-machine model keeps your material on this machine with semantic results. A remote endpoint receives every text that is embedded.
• Before the first send to a remote endpoint in a session, you are shown the address and how much text is going. Changing the address asks again.
• Nothing is embedded automatically, not on save and not on drop.
• Changing the model or the vector length starts a separate index. The old one stays on disk until you clear it, and no search mixes the two.
• A search that fell back to keywords is labelled as such, never shown as semantic.