# Config `knowledge.retrieval` in `daemon.yaml` configures the retrieval engine: the qmd searcher and the query rewriter. The [method & delivery](/knowledge-base/modes) knobs decide how results reach an agent; this block decides how results are produced — its settings apply to retrieving pipelines (`method: bm25`/`vector`), `bento ask`, and the `ask_knowledge` MCP tool alike. The engine initializes whenever knowledge sources are discovered and `qmd` is on PATH. There is no enable switch — to keep retrieval out of pipeline runs, omit `knowledge:` blocks (and `defaults.knowledge`), or set `delivery: none`. ```yaml knowledge: retrieval: topK: 20 # default 5 rewrite: target: claude-fast ``` | Field | Default | Description | |-------|---------|-------------| | `topK` | `5` | Default results per source. Raise to `20` for small knowledge bases (fewer than ~20 docs); `ask_knowledge` always uses this value. Per-run `knowledge.topK` overrides it. | | `minScore` | `0` | Drop results scoring below this. | | `rewrite.target` | — | Named execution target. Omit to search with the raw prompt. | | `rewrite.timeoutMs` | `15000` | Hard cap on the rewriter spawn; on timeout the raw prompt is searched as-is. | ## rewrite When `rewrite.target` is set, the daemon spawns a one-shot CLI call that condenses the prompt into at most three queries of 3–6 words each, preserving proper nouns, file names, and error strings. Each query is searched; results merge. The rewriter serves every daemon retrieval: `method: bm25` / `vector` pipelines, the daemon-backed `bento ask`, and `ask_knowledge`. ## Per-run settings Injection behavior — mode, filter, instructions, `topK`, `maxTokens` (budget, default `8000`) — lives in the per-pipeline [`knowledge:` block](/pipelines/config#knowledge) and `defaults.knowledge`. Engine settings here apply underneath: `rewrite.*` and `minScore` always, `topK` unless the run overrides it.