Search

The search box on the document list finds documents by their contents, not just their titles. This page covers how it matches and what it returns.

How matching works

Search combines two methods and blends their results:

• Full-text matching finds documents whose text matches your search words, and highlights matching snippets. Each section of a document is indexed with a short generated description of its context, so a section can also match words that describe it rather than appear in it verbatim.
• Semantic matching finds documents that are about what you searched for, even when they use different wording. A result found only this way shows a plain excerpt of the best-matching passage rather than a highlighted snippet.

Blending the two means you can find a document by the exact words in it or by its topic. If semantic matching is not available on the server, search falls back to full-text matching on its own.

Semantic matching works by computing embeddings of your document text with Google's embedding API, and the per-section context descriptions are generated by an AI model. See the privacy policy for how document content is shared with AI providers.

What you can search

Search covers the title and body of each document. Results are always limited to documents you can access: search never surfaces a document you do not have a role on.

The document list search reads document contents. External agents can run the same content search through the list_documents tool. See Agents and MCP.

Keeping the index current

After you edit a document, its search entry updates on its own about half a minute after you stop editing, so searches reflect recent edits without any action from you.