harvest-code 1.5.0

Harvest codebases into portable JSON + chunks for RAG and tooling

harvest-code

Harvest codebases into portable JSON + chunks for RAG, tooling, and analysis.

Quick Start

pip install harvest-code
harvest serve  # Watches current dir, auto-creates harvest, serves at http://localhost:8787

Interactive web interface with search, filtering, syntax highlighting, and progressive loading

Key Features

• 🔍 Zero-dependency codebase harvesting from local directories or GitHub URLs
• 📊 Smart chunking by functions, classes, exports with content-agnostic IDs
• 🌐 Interactive Web UI with search, filtering, syntax highlighting, and deep linking
• 🚀 Progressive loading - handles 50k+ files with infinite scroll pagination
• 📝 Schema v1.2 with stable chunk identifiers for incremental updates
• 🎯 Intelligent defaults - automatically skips binaries, caches, build artifacts, and lockfiles
• ⚡ Fast filtering - by language, visibility, file patterns, and symbols
• 💾 Saved views - bookmark and share specific filtered states

Use Cases

• RAG/AI Systems: Feed structured code context to LLMs with precise chunk boundaries
• Documentation: Auto-generate API references from exported symbols and functions
• Code Analysis: Extract dependencies, exports, and architectural patterns
• Legacy Migration: Understand large codebases through progressive exploration
• Code Search: Find functions, classes, and patterns across entire repositories

Installation

pip install harvest-code

Requirements: Python 3.9+ (no dependencies)

Harvest file naming & location

• Canonical extension: .harvest.json
• Default location: ./codebase.harvest.json (in current directory)
• The harvester and watcher automatically ignore any *.harvest.json files to prevent recursive harvesting
• Legacy filenames (.hvst.json, .har.json, .harvest-code.json) are still supported but the canonical name is recommended

Tip: choose whether to commit the index:

# .gitignore (optional - exclude harvest files from git)
.harvest/

# OR commit them for reproducible runs
git add .harvest/codebase.harvest.json

Core Commands

Harvest Local Directory

# Basic harvesting
harvest reap .  # Creates ./codebase.harvest.json
harvest reap /path/to/project  # Creates /path/to/project/codebase.harvest.json

# Custom output path
harvest reap . -o my-custom-harvest.json

# Control what's included
harvest reap . --include data          # Only file contents, no chunks
harvest reap . --include metadata data  # Metadata + data, no chunks  
harvest reap . --exclude chunks        # Everything except chunks
harvest reap . --include chunks         # Only chunks (for analysis)

# With size limits
harvest reap . --max-files 1000 --max-bytes 512000

# Include everything (override smart defaults)
harvest reap . --no-default-excludes -o full-harvest.json

Harvest from GitHub

harvest reap https://github.com/user/repo -o repo.harvest.json
harvest reap https://github.com/user/repo/tree/main/src -o src-only.harvest.json

Query & Filter

# Find all public Python functions
harvest query output.json --entity chunks --language python --public true

# Search specific paths with custom fields
harvest query output.json --path-glob "src/**" --fields path,symbol,kind,start_line

# Find exports and components
harvest query output.json --entity files --export-named "MyComponent" 
harvest query output.json --entity chunks --kind export_default --language typescript

Interactive Web Interface

harvest serve                    # Watch + serve current directory (default)
harvest serve /path/to/project    # Watch + serve specific directory
harvest serve --no-watch          # Serve without watching for changes
harvest serve --port 8080         # Use custom port

Web UI Features:

• Search & Filter: Real-time filtering by language, path, symbols
• Syntax Highlighting: Code highlighting enabled by default with dropdown toggle (Highlight/Plain)
• Auto-refresh: Seamless updates when files change in watch mode
• Progressive Loading: Infinite scroll for large datasets
• Deep Linking: Share URLs to specific files and line ranges
• Saved Views: Bookmark filtered states for quick access
• Mobile Responsive: Works on desktop, tablet, and mobile

Live Updates (auto-refresh, zero deps)

harvest serve now includes watching by default - no need for separate terminals!

# Single command for watch + serve (recommended)
harvest serve

# Or run them separately if needed
harvest watch .                   # Terminal 1: Watch only
harvest serve --no-watch          # Terminal 2: Serve only

The UI polls /api/meta every 3 seconds and automatically refreshes when files change:

• Updates the file list and metadata instantly
• Refreshes the content preview without manual clicking
• Maintains your scroll position and selection
• Shows progress in browser console ([harvest] logs) for debugging

All API endpoints send Cache-Control: no-store headers to prevent stale data, and the UI includes version parameters for cache busting.

New endpoint:

GET /api/harvest                # returns the current JSON, uncached, with ETag

Watch Flags:

• --debounce-ms 800 - Coalesce bursts of file events (milliseconds)
• --poll 1.0 - Filesystem snapshot cadence (seconds)
• --only-ext py,ts,js - Include only specific extensions
• --skip-ext log,tmp - Exclude specific extensions

The watcher uses portable polling that works on all platforms and performs incremental re-harvesting when files change.

Output Control

You can control what's included in the harvest output:

• --include metadata data - Just files and metadata (faster, smaller)
• --include data - Just file contents (minimal output)
• --exclude chunks - Skip chunk generation (useful for simple file archiving)
• --include chunks - Just chunks (for code analysis without full content)

Common use cases:

# Fast file inventory without content
harvest reap . --include metadata

# Full content without chunking overhead  
harvest reap . --exclude chunks

# Just chunks for LLM context
harvest reap . --include chunks --format jsonl

Output Schema

harvest-code produces JSON with three main sections:

{
  "metadata": {
    "schema": "harvest/v1.2",
    "source": {"type": "local", "root": "/path/to/code"},
    "counts": {"total_files": 150, "total_bytes": 524288},
    "created_at": "2025-08-15T12:00:00Z"
  },
  "data": [
    {
      "path": "src/utils.py",
      "language": "python", 
      "size": 1024,
      "exports": null,
      "py_symbols": {"functions": ["helper"], "classes": ["Util"]},
      "content": "def helper():\n    pass\n\nclass Util:\n    pass"
    }
  ],
  "chunks": [
    {
      "id": "abc123...",
      "file_path": "src/utils.py",
      "language": "python",
      "kind": "function",
      "symbol": "helper", 
      "start_line": 1,
      "end_line": 2,
      "public": true
    }
  ]
}

Advanced Usage

Custom Exclusions

# Skip additional file types
harvest reap . --skip-ext .log,.tmp --skip-folder cache,temp

# Include only specific languages  
harvest reap . --only-ext .py,.js,.ts

Incremental Updates

# Reuse unchanged files for faster re-harvesting
harvest reap . --prev previous-harvest.json -o updated.json

Large Codebases

The web UI automatically handles large datasets with:

• Pagination: Loads 50 items at a time by default
• Infinite Scroll: Seamlessly loads more results
• Cursor-based Navigation: Efficient browsing of 10k+ items
• Smart Filtering: Client-side filtering for responsive UX

Default Exclusions

harvest-code intelligently skips common non-source files:

File Extensions: .log, .tmp, .db, .sqlite, .mp4, .jpg, .png, .zip, .exe, .pyc, .min.js, .woff, .pb, .tflite, etc.

Directories: node_modules, .git, dist, build, __pycache__, .pytest_cache, .mypy_cache, .vscode, .idea, .next, Pods, DerivedData, etc.

Lockfiles: yarn.lock, package-lock.json, poetry.lock, Cargo.lock, go.sum, etc.

Use --no-default-excludes to include everything.

Language Support

Chunking & Symbol Extraction:

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

Syntax Highlighting: Python, JavaScript, TypeScript, JSON, YAML, TOML, Markdown, Shell, Go, Rust

API Integration

The web server exposes REST endpoints for programmatic access:

curl http://localhost:8787/api/search?entity=chunks&language=python&limit=10
curl http://localhost:8787/api/meta  # Get harvest metadata

Examples

RAG Pipeline Integration

import json

# Load harvest data
with open('codebase.json') as f:
    harvest = json.load(f)

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

Documentation Generation

# Extract all exported React components
harvest query react-app.json --entity chunks --kind export_default --language typescript --fields symbol,file_path > components.txt

Migration Analysis

# Find all Python classes for refactoring
harvest query legacy-app.json --entity chunks --kind class --language python --min-lines 10

Contributing

We welcome contributions! Please see our Contributing Guidelines for details.

Development Setup:

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

License

MIT License - see LICENSE file for details.

Links

• PyPI: https://pypi.org/project/harvest-code/
• GitHub: https://github.com/veyorokon/code-harvest
• Issues: https://github.com/veyorokon/code-harvest/issues
• Documentation: https://github.com/veyorokon/code-harvest/wiki