active tools

src2md

Convert source code to structured, context-optimized markdown for LLMs with intelligent summarization.

Started 2025 Python

Resources & Distribution

Source Code

GitHub Repository

Package Registries

PyPI
1 Stars

src2md

License PyPI Python Versions Version

src2md is a powerful tool that converts source code repositories into structured, context-window-optimized representations for Large Language Models (LLMs). It addresses the fundamental challenge of fitting meaningful codebases into limited context windows while preserving the most important information through intelligent summarization, AST-based analysis, and optional LLM-powered compression.

🚀 Features

New in v2.1

  • 🎯 Context Window Optimization: Intelligently fit codebases into LLM context windows with smart truncation
  • 📝 Intelligent Summarization: AST-based code analysis with multiple compression levels
  • 🤖 LLM-Powered Compression: Optional OpenAI/Anthropic integration for semantic summarization
  • ⚡ Fluent API: Elegant method chaining with new summarization methods
  • 📊 File Importance Scoring: Multi-factor analysis to prioritize critical files
  • 🪟 Predefined LLM Windows: Built-in support for GPT-4, Claude, and more
  • 🔄 Progressive Summarization: Multi-tier compression strategies for different file types

Core Features

  • Multiple Output Formats: JSON, JSONL, Markdown, HTML, and plain text
  • Smart Token Management: Accurate token counting with tiktoken and structure-aware truncation
  • Multi-Language Support: Specialized summarizers for Python, JavaScript, TypeScript, JSON, YAML
  • Code Statistics: Automatic generation of project metrics and complexity analysis
  • Flexible Filtering: Customizable include/exclude patterns
  • Rich CLI Interface: Beautiful progress indicators and colored output

📦 Installation

Install via PyPI using pip:

pip install src2md

🛠️ Usage

Quick Start - Fluent API

from src2md import Repository, ContextWindow

# Basic usage
output = Repository("/path/to/project").analyze().to_markdown()

# Optimize for GPT-4 context window
output = (Repository("/path/to/project")
    .optimize_for(ContextWindow.GPT_4)
    .analyze()
    .to_markdown())

# Full fluent API with all features
result = (Repository("/path/to/project")
    .name("MyProject")
    .branch("main")
    .include("src/", "lib/")
    .exclude("tests/", "*.log")
    .with_importance_scoring()
    .with_summarization(
        compression_ratio=0.3,  # Target 30% of original size
        preserve_important=True,  # Keep critical files intact
        use_llm=True  # Use LLM if available
    )
    .prioritize(["main.py", "core/"])
    .optimize_for_tokens(100_000)  # 100K token limit
    .analyze()
    .to_json(pretty=True))

Command Line Interface

# Basic markdown generation
src2md /path/to/project -o documentation.md

# With context optimization
src2md /path/to/project --gpt4 -o optimized.md
src2md /path/to/project --claude3 --importance

# With intelligent summarization
src2md /path/to/project --summarize --compression-ratio 0.3
src2md /path/to/project --summarize-tests --summarize-docs

# With LLM-powered summarization (requires API key)
src2md /path/to/project --use-llm --llm-model gpt-3.5-turbo

# Multiple output formats
src2md /path/to/project --format json --pretty
src2md /path/to/project --format html -o docs.html

Python API Examples

Basic Context Optimization

from src2md import Repository, ContextWindow

# Optimize for different LLM context windows
repo = Repository("./my-project")
output = repo.optimize_for(ContextWindow.CLAUDE_3).analyze().to_markdown()

# Custom token limit with importance scoring
repo = (Repository("./my-project")
    .with_importance_scoring()
    .optimize_for_tokens(50_000)
    .analyze())

Intelligent Summarization

# Enable smart summarization with compression
repo = (Repository("./my-project")
    .with_summarization(
        compression_ratio=0.3,  # Compress to 30% of original
        preserve_important=True,  # Keep critical files intact
        use_llm=False  # Use AST-based summarization
    )
    .optimize_for(ContextWindow.GPT_4)
    .analyze())

# Use LLM-powered summarization (requires API key)
import os
os.environ['OPENAI_API_KEY'] = 'your-key-here'

repo = (Repository("./my-project")
    .with_summarization(
        compression_ratio=0.2,  # More aggressive compression
        use_llm=True,
        llm_model="gpt-3.5-turbo"
    )
    .analyze())

Multi-Tier Compression Strategy

# Configure different summarization levels for different file types
repo = (Repository("./my-project")
    .with_importance_scoring()
    .prioritize(["src/core/", "api/"])  # Critical paths
    .summarize_tests()  # Compress test files
    .summarize_docs()   # Compress documentation
    .with_summarization(
        compression_ratio=0.25,
        preserve_important=True
    )
    .optimize_for_tokens(100_000)
    .analyze())

# Access summarization metadata
data = repo.to_dict()
for file in data['source_files']:
    if file.get('was_summarized'):
        print(f"Summarized {file['path']}: {file['original_size']} -> {file['size']} bytes")

Generate Multiple Formats

repo = Repository("./my-project").analyze()
markdown = repo.to_markdown()
json_data = repo.to_json()
html_doc = repo.to_html()

# Access raw data
data = repo.to_dict()
print(f"Files: {data['metadata']['file_count']}")
print(f"Token usage: {data['metadata'].get('total_tokens', 0)}")
print(f"Compression achieved: {data['metadata'].get('compression_ratio', 1.0):.1%}")

🎯 Summarization Features

AST-Based Python Summarization

src2md uses Abstract Syntax Tree (AST) analysis to intelligently summarize Python code while preserving structure:

  • MINIMAL: Only class/function signatures
  • OUTLINE: Signatures with structural hierarchy
  • DOCSTRINGS: Signatures plus documentation
  • SIGNATURES: Full signatures with type hints
  • FULL: No summarization

Multi-Language Support

Specialized summarizers for different file types:

  • Python: AST-based analysis with import/export preservation
  • JavaScript/TypeScript: Function and class extraction
  • JSON/YAML: Schema extraction with sample data
  • Test Files: Test name and assertion extraction
  • Documentation: Heading and key point extraction

Smart Truncation

When files must be truncated to fit token limits:

  1. Preserves code structure (complete functions/classes)
  2. Maintains syntax validity
  3. Prioritizes public APIs over private methods
  4. Keeps imports and exports intact

LLM-Powered Summarization

Optional integration with OpenAI and Anthropic for semantic compression:

# Set API keys
export OPENAI_API_KEY="your-openai-key"
export ANTHROPIC_API_KEY="your-anthropic-key"

# Use LLM summarization
src2md /path/to/project --use-llm --llm-model gpt-3.5-turbo
src2md /path/to/project --use-llm --llm-model claude-3-haiku-20240307

📊 Output Formats

JSON

Structured data perfect for programmatic processing:

{
  "metadata": {
    "project_name": "my-project",
    "generated_at": "2025-01-01T12:00:00",
    "patterns": {...}
  },
  "statistics": {
    "total_files": 42,
    "languages": {"python": {"count": 15, "total_size": 50000}},
    "project_complexity": 3.2
  },
  "documentation": [...],
  "source_files": [...]
}

JSONL

One JSON object per line - perfect for streaming and big data tools:

{"type": "metadata", "data": {...}}
{"type": "statistics", "data": {...}}
{"type": "source_file", "data": {...}}

HTML

Beautiful, styled documentation ready for the web with syntax highlighting and responsive design.

Markdown

Clean, readable documentation compatible with GitHub, GitLab, and other platforms.

🔧 Advanced Options

File Patterns

# Custom documentation patterns
src2md project --doc-pat '*.md' '*.rst' '*.txt'

# Specific source file types
src2md project --src-pat '*.py' '*.js' '*.ts'

# Ignore patterns
src2md project --ignore-pat '*.pyc' 'node_modules/' '.git/'

Ignore Files

Create a .src2mdignore file in your project root:

# Dependencies
node_modules/
__pycache__/
*.pyc

# Build outputs
dist/
build/
*.egg-info/

# IDE files
.vscode/
.idea/

Configuration

# Use custom ignore file
src2md project --ignore-file .gitignore

# Disable statistics
src2md project --no-stats

# Metadata only (no file contents)
src2md project --no-content

🎯 Use Cases

  • LLM Context: Generate structured context for AI/ML models
  • Documentation: Create beautiful project documentation
  • Code Analysis: Extract metrics and statistics from codebases
  • Data Export: Convert code to structured formats for analysis
  • Archive: Create comprehensive snapshots of projects
  • CI/CD: Generate documentation automatically in build pipelines

📈 Statistics & Metrics

src2md automatically generates:

  • File Metrics: Counts by type and language
  • Code Complexity: Cyclomatic complexity scores
  • Token Usage: Actual token counts for LLM context
  • Compression Stats: Before/after summarization metrics
  • Importance Scores: File prioritization rankings
  • Language Breakdown: Distribution of code by language
  • Structure Analysis: Dependency and module relationships

🤝 Migration from v0.x

The new version is backward compatible. Existing commands work unchanged:

# This still works exactly as before
src2md project -o docs.md --doc-pat '*.md' --src-pat '*.py'

New features are opt-in through additional flags and the Python API.

📄 License

MIT License - see LICENSE file for details.

Discussion