Find which files and lines most likely contain what you describe — the cause of a bug from its symptoms, the code implementing a feature, where a setting is defined. Pass 1: every file is split into chunks and ALL chunks are scored by the decision model (nothing is truncated; very cheap even for megabytes). Pass 2: the best chunks are re-scored in small line windows to pinpoint the lines. Returns ranked file:line ranges with probabilities, a per-file ranking and the code of the top locations, so you can go straight to the fix.

Classify every file in the given paths into your labels with a decision model (one very cheap request per file, calibrated confidence). Returns files grouped by label. Good for triage: tests/configs/dead code, sorting documents by topic, flagging files about X. Label descriptions after a colon noticeably improve accuracy.

Ask several typed questions about EACH file with a decision model: yes_no flags, one-of choices and ordered scores, with calibrated probabilities. Example: for every file decide uses_db (yes_no), layer (choice: api/domain/ui/infra), complexity (score: low/medium/high). Use only_matching to return just the files where a yes_no question is yes.

List the decision models available on OpenRouter (id, context window, input price) and which one is the default. Pass an id as model to any other tool. Bigger context = bigger chunks per request.