harvest-code 1.6.0

Harvest codebases into portable JSON + chunks for RAG and tooling

harvest-code

Extract codebases into portable JSON for RAG, tooling, and analysis.

Quick Start

pip install harvest-code
harvest serve  # Watch + serve current directory at http://localhost:8787

Features

• 🔍 Zero dependencies - Pure Python, no external packages required
• 📊 Smart chunking - Extracts functions, classes, exports with stable IDs
• 🌐 Interactive UI - Web interface with search, filtering, and syntax highlighting
• 🎯 Intelligent filtering - Automatically skips build artifacts, binaries, and test files
• ⚡ Live updates - Auto-refreshes when files change (watch mode)
• 🚀 Scales - Handles 50k+ files with progressive loading

Commands

harvest reap - Extract codebase to JSON

# Basic usage - output named after directory
harvest reap .                    # → ./current-dir-name.harvest.json
harvest reap /path/to/project     # → ./project.harvest.json

# Custom output
harvest reap . -o analysis.json

# Control what's included
harvest reap . --include metadata       # File inventory only
harvest reap . --include data          # File contents only  
harvest reap . --exclude chunks        # Skip code parsing
harvest reap . --format jsonl          # Line-delimited JSON

# Override filters
harvest reap . --no-default-excludes   # Include hidden files, tests, etc.

harvest serve - Interactive web UI

# Watch + serve (default)
harvest serve                    # Current directory on port 8787
harvest serve /path/to/project   # Specific directory
harvest serve --port 8080        # Custom port

# Control options
harvest serve --no-watch         # Disable auto-refresh
harvest serve --only-ext py,ts   # Watch specific file types

Web UI Features:

• Real-time search and filtering
• Syntax highlighting with toggle
• Auto-refresh on file changes
• Infinite scroll for large codebases
• Deep linking to specific files/lines

harvest query - Search harvest data

# Find specific code elements
harvest query data.json --entity chunks --language python --public true
harvest query data.json --entity files --export-named "MyComponent"
harvest query data.json --path-glob "src/**" --fields path,symbol,kind

harvest watch - Monitor directory changes

# Continuous harvesting
harvest watch .                        # → ./<dirname>.harvest.json
harvest watch /path/to/src -o out.json # Custom output
harvest watch . --only-ext py,js,ts    # Specific extensions

harvest sow - Generate artifacts

# Create React barrel exports
harvest sow data.json --react src/index.ts

Output Structure

harvest generates JSON with three sections:

{
  "metadata": {
    "schema": "harvest/v1.2",
    "source": {"type": "local", "root": "/path/to/code"},
    "counts": {"total_files": 150, "total_bytes": 524288}
  },
  "data": [
    {
      "path": "src/utils.py",
      "language": "python",
      "content": "def helper():\n    pass",
      "exports": null,
      "py_symbols": {"functions": ["helper"]}
    }
  ],
  "chunks": [
    {
      "id": "abc123...",
      "file_path": "src/utils.py",
      "kind": "function",
      "symbol": "helper",
      "start_line": 1,
      "end_line": 2,
      "public": true
    }
  ]
}

Output Sections

• metadata - File inventory, counts, timestamps
• data - File contents and language metadata
• chunks - Parsed symbols (functions, classes, exports)

Control output with --include and --exclude:

harvest reap . --include metadata       # Inventory only (fast)
harvest reap . --exclude chunks         # Skip parsing (smaller)
harvest reap . --include chunks         # Symbols only (for analysis)

File Handling

Smart Filtering

harvest uses a three-tier filtering system:

Completely Skipped:

• Hidden files and directories (.git/, .env)
• Test directories (tests/, __tests__/)
• Build artifacts (dist/, build/, node_modules/)
• Binaries and media (.exe, .mp3, .db)
• Logs and temp files (.log, .tmp)

Path-Only (no content):

• Images (.jpg, .png, .svg)
• Fonts (.ttf, .woff)
• Documents (.pdf, .doc)

Fully Processed:

• Source code (.py, .js, .ts, etc.)
• Config files (.json, .yaml, .toml)
• Documentation (.md, .txt)

Override with --no-default-excludes to include everything.

Language Support

Full parsing (chunks + symbols):

• Python - Functions, classes via AST
• JavaScript/TypeScript - Functions, classes, React components, ES6/CommonJS exports
• JSON/YAML/TOML - Single file chunks

Syntax highlighting in web UI: Python, JavaScript, TypeScript, JSON, YAML, TOML, Markdown, Shell, Go, Rust

API Endpoints

The web server exposes REST APIs:

# Search chunks
curl http://localhost:8787/api/search?entity=chunks&language=python

# Get metadata
curl http://localhost:8787/api/meta

# Download full harvest
curl http://localhost:8787/api/harvest

Use Cases

RAG/LLM Context

import json

with open('project.harvest.json') as f:
    harvest = json.load(f)

# Extract public functions for context
api_functions = [
    chunk for chunk in harvest['chunks'] 
    if chunk['public'] and chunk['kind'] == 'function'
]

Code Analysis

# Find all React components
harvest query app.json --entity chunks --kind export_default --language typescript

# Find large classes
harvest query app.json --entity chunks --kind class --min-lines 100

Documentation Generation

# Extract all exports
harvest query app.json --entity files --has-default-export --fields path,exports

File Naming

• Output files use pattern: <directory-name>.harvest.json
• Created in current working directory (where command is run)
• Use -o flag for custom names/paths
• All *.harvest.json files are automatically ignored to prevent recursion

Development

git clone https://github.com/veyorokon/code-harvest.git
cd code-harvest
pip install -e .
harvest reap . -o self.harvest.json

Links

• PyPI Package
• GitHub Repository
• Issue Tracker

License

MIT - see LICENSE file