hygrep 0.0.1

Hybrid grep: fast scanning + neural reranking

hygrep

Hybrid grep: fast scanning + neural reranking

pip install hygrep
hygrep "auth logic" ./src

What it does

• Semantic search: "auth" finds "login", "session", "token"
• Smart context: Returns full functions/classes, not just lines
• Fast: Parallel regex recall (~20k files/sec), then neural reranking
• Zero indexing: Works instantly on any codebase

Install

pip install hygrep
# or
uv tool install hygrep
# or
pipx install hygrep

First search downloads the model (~83MB, cached in ./models/).

Usage

hygrep "query" [path]           # Search (default: current dir)
hygrep "error handling" . -n 5  # Limit to 5 results
hygrep "auth" . --fast          # Skip reranking (instant grep)
hygrep "test" . -t py,js        # Filter by file type
hygrep "config" . --json        # JSON output for agents/scripts
hygrep info                     # Check installation status

Run hygrep --help for all options.

Output

src/auth.py:42 [function] login (0.89)
src/session.py:15 [function] validate_token (0.76)

With --json:

[{"file": "src/auth.py", "type": "function", "name": "login", "start_line": 42, "score": 0.89, "content": "def login(user): ..."}]

Config

Optional ~/.config/hygrep/config.toml:

n = 10
color = "always"
exclude = ["*.test.js", "tests/*"]

Supported Languages

Python, JavaScript, TypeScript, Go, Rust, C, C++, Java, Ruby, C#, Mojo

How it works

Query → Parallel regex scan → Tree-sitter extraction → ONNX reranking → Results

Stage	What
Recall	Mojo/Python parallel scanner (~20k files/sec)
Extract	Tree-sitter AST (functions, classes)
Rerank	ONNX cross-encoder (mxbai-rerank-xsmall-v1)

Development

git clone https://github.com/nijaru/hygrep && cd hygrep
pixi install && pixi run build-ext && pixi run test

License

MIT