wiki-query

You query an internal wiki by consulting an LLM-maintained index to grep relevant terms, load matching pages, and generate answers grounded with strict wikilink citations. This prevents model hallucination by checking existing knowledge first.

The best way to structure a knowledge base search uses a wiki-miss flow that derives an answer from model or web knowledge when the wiki lacks information, then optionally proposes a new wiki page to capture that derived answer for future reuse.

To add citations to answers from a local knowledge management system, the system loads matched wiki pages during retrieval and enforces strict [[wikilink]] citation integrity, linking every grounded claim directly back to its source page.

Yes, you can use grep-based indexing to search a markdown wiki. The system greps the wiki index file for relevant terms before scanning the rest of the wiki, only falling back to web or MCP search if the wiki lacks the needed information.

You should avoid using a wiki lookup tool for entirely novel queries where no prior discussions, decisions, or concepts exist in the wiki, as the system will bypass local search and immediately fall back to web or model knowledge.

Yes, wiki-query works with MCP search tools. When the local wiki misses a query, the system falls back to web, MCP, or model knowledge to derive an answer and optionally proposes a new wiki page for future reuse.