hhg 0.0.25

Hybrid file search — semantic + keyword matching

hhg (hybrid grep)

Hybrid file search — semantic + keyword matching

pip install hhg
hhg build ./src
hhg "authentication flow" ./src

What it does

Search code and text using natural language. Combines semantic understanding with keyword matching (BM25) for accurate results:

$ hhg build ./src                    # Build index first
Found 40 files (0.0s)
✓ Indexed 646 blocks from 40 files (34.2s)

$ hhg "error handling" ./src         # Then search
api_handlers.ts:127 function errorHandler
  function errorHandler(err: Error, req: Request, res: Response, next: NextFunc...

errors.rs:7 class AppError
  pub enum AppError {
      Database(DatabaseError),

2 results (0.52s)

Why hhg over grep?

grep finds exact text. hhg understands what you're looking for.

Query	grep finds	hhg finds
"error handling"	Comments mentioning it	errorHandler(), AppError
"authentication"	Strings containing "auth"	login(), verify_token()
"database"	Config files, comments	Connection, query(), Db

Hybrid search combines semantic understanding (finds related concepts) with BM25 keyword matching (finds exact terms). Best of both worlds.

Use grep/ripgrep for exact strings (TODO, FIXME, import statements). Use hhg when you want implementations, not mentions.

Install

Requires Python 3.11-3.13 (onnxruntime lacks 3.14 support).

pip install hhg
# or
uv tool install hhg
# or
pipx install hhg

The embedding model (jina-code-int8) downloads on first use (~154MB).

Usage

hhg build [path]                # Build/update index (required first)
hhg "query" [path]              # Semantic search
hhg status [path]               # Check index status
hhg list [path]                 # List all indexes under path
hhg clean [path]                # Delete index
hhg clean [path] -r             # Delete index and all sub-indexes

# Options
hhg -n 5 "error handling" .     # Limit results
hhg --json "auth" .             # JSON output for scripts/agents
hhg -l "config" .               # List matching files only
hhg -t py,js "api" .            # Filter by file type
hhg --exclude "tests/*" "fn" .  # Exclude patterns
hhg --exclude "*.md" "api" .   # Code only (exclude docs)

# Model
hhg model                       # Check if model is installed
hhg model install               # Download model (auto-downloads on first use)

Note: Options go before positional args, or use -- separator:

hhg --exclude "*.md" "api" .   # Options first
hhg "api" . -- --exclude "*.md" # Or use -- separator

Output

Default:

src/auth.py:42 function login
  def login(user, password):
      """Authenticate user and create session."""
      ...

JSON (--json):

[
  {
    "file": "src/auth.py",
    "type": "function",
    "name": "login",
    "line": 42,
    "end_line": 58,
    "content": "def login(user, password): ...",
    "score": 0.87
  }
]

Compact JSON (--json --compact): Same fields without content.

How it Works

Query → Embed → Hybrid search (semantic + BM25) → Results
                        ↓
             Requires 'hhg build' first (.hhg/)
             Auto-updates stale files on search

Supported Files

Code (22 languages): Bash, C, C++, C#, Elixir, Go, Java, JavaScript, JSON, Kotlin, Lua, Mojo, PHP, Python, Ruby, Rust, Svelte, Swift, TOML, TypeScript, YAML, Zig

Text: Markdown, plain text, RST — smart chunking with header context for docs, blog posts, research papers

Development

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

License

MIT