Smarter File Search: Give Your Assistant the Right Information
File summaries, outlines, and meaning-based search help agents use the context you have collected.
By David Hemphill ·
Files now give the assistant more ways to find the material it needs. Each processed file gets a short summary, and an outline tool lets an agent inspect its sections before reading a particular part.
Search also combines words with meaning where embeddings are available. A request about refund terms can find a passage described as cancellation policy, even when the vocabulary differs. Exact searches remain exact: quoted phrases and excluded words use keyword search, and keyword search remains the fallback when semantic search is unavailable.
For a project with several interviews or reference documents, that means you can ask about a decision without first knowing which file contains it. The assistant can discover the source and read the relevant part rather than treat every file as an undifferentiated block of text.
Connected agents can also create and update text files over MCP, useful for keeping a README or reference note current. Those file-writing tools are for external connections, not the in-app assistant, and do not overwrite extracted PDF or image text.
Open Livemark and ask a question about your project's source files.