QueryFind is a local macOS-first CLI for finding files with natural-language queries.
It is useful when:
- you know the information you want, but not the exact file name
- the answer is likely inside file contents, not just in paths
- you want a constrained local search agent instead of arbitrary shell execution
QueryFind uses a local Ollama model plus a read-only command allowlist (fd, rg, ls, tree, stat, mdls) to plan searches, inspect results, and return ranked matches with evidence.
Requirements:
- macOS
- Python 3.12+
fdrgtreerecommendedollama- a local Ollama model, for example
qwen3.5:27borqwen3-coder:30b
Typical setup with Homebrew:
brew install fd ripgrep tree ollama
python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
ollama pull qwen3-coder:30bCheck the environment:
python -m queryfind --doctor
make testRun a search:
python -m queryfind "find the latest signed contract for Redwood"Search a specific folder:
python -m queryfind "find the Wi-Fi password for Harbor 7" --root ~/DocumentsRun without the LLM:
python -m queryfind "find project files about Atlas" --no-llmUseful commands:
make doctor
make synthetic-eval
python -m queryfind.benchmark --model qwen3-coder:30b
python -m queryfind.benchmark --manifest benchmark_fs/extended_manifest.json --model qwen3-coder:30b
python -m queryfind.benchmark --manifest benchmark_fs/handmade100_manifest.json --model qwen3-coder:30b- QueryFind auto-starts Ollama when possible.
- LLM mode is local-only and read-only by default. Non-local Ollama URLs are rejected unless
--allow-remote-ollamais passed explicitly. - Backend command execution is constrained to trusted macOS binaries plus validated read-only argv shapes.
- Heuristic mode is available explicitly with
--no-llm. - Logs are written under
.queryfind/.