Documentation index
Index a library's documentation site or an internal docs folder on this computer so the agent searches it with search_docs and answers with sources.
What it does
The agent does not know APIs or internal rules that changed after it was trained. Index the documentation in the Docs tool and the agent, instead of guessing, finds the passage with the search_docs tool and answers with the source URL or file path. The list of indexed sets also goes into the system prompt, so the agent knows when to look.
Adding documentation
- Add site: enter the docs start address (for example
https://docs.example.com/guide/). Only pages on the same site under the same path are fetched, up to 200 pages. The site'srobots.txt(Disallow, Crawl-delay) is honoured. - Add folder: choose an internal docs folder. Up to 2,000
.md,.mdx,.txt,.rst,.adocand.htmlfiles (1 MB each) are read; folders such asnode_modules,.gitanddistare skipped.
Progress (pages, chunks) shows while indexing and the set turns "Ready" when done. Press Re-index when the docs change.
How search works
Pages are split into chunks of about 1,200 characters and ranked with BM25. Korean text is also matched by two-character pieces inside words, so different spacing still matches. No embedding model or server is involved, so there is no cost and both indexing and search run on this computer. The index is stored in ~/.level11/docs.
Test search at the bottom shows what the agent would receive.
Turning it off per agent
In the agent editor, untick "May search the indexed documentation with search_docs" and that agent does not get the docs search tool.
Limits
- Documentation sites that render their content only with JavaScript may index as empty pages. Index the repository's docs folder instead.
- Pages behind a sign-in cannot be fetched.