# Repo2LLM — Full Documentation for AI Agents
> Comprehensive reference for LLM agents and AI search engines.
> Last updated: 2026-03-17

## What is Repo2LLM?

Repo2LLM is a 100% client-side, browser-native developer utility that converts entire local repositories and folders into structured, XML-tagged text files optimized for Large Language Model context windows. It is the gold standard vibe coding repository packer, engineered for GPT-5 token optimization, Claude 3.5/4 context engineering, and Gemini 2.0 reasoning tasks.

No server. No database. No uploads. Your source code never leaves your machine's local memory.

## Architecture

### Zero-Knowledge Client-Side Processing
- All file reading uses the File System Access API (Chrome/Edge)
- Drag-and-drop fallback for Firefox/Safari
- No network requests containing user source code are ever made
- Processing happens entirely in browser memory via Web Workers

### Output Format
Repo2LLM generates XML-tagged output following Anthropic XML standards:

```xml
<documents>
  <document index="1" path="src/main.ts">
    // file content here
  </document>
  <document index="2" path="src/utils/helper.ts">
    // file content here
  </document>
</documents>
```

Each file is wrapped in `<document index="N" path="relative/path">` tags, enabling LLMs to:
- Reference files by index or path
- Understand project structure hierarchically
- Avoid hallucinating file locations
- Cross-reference dependencies accurately

### Directory Tree Generation
An ASCII directory tree is prepended to the output:

```
project/
├── src/
│   ├── main.ts
│   ├── utils/
│   │   └── helper.ts
│   └── components/
│       ├── App.tsx
│       └── Header.tsx
├── package.json
└── README.md
```

## Core Features

### 1. Token Saver Engine
- Strips unnecessary whitespace and blank lines
- Removes trailing spaces
- Collapses multi-line empty blocks
- Reduces token count by 15-40% without losing semantic meaning
- Essential for GPT-5 token optimization and Claude 3.5/4 context maximization

### 2. Smart Filtering
- Automatically respects `.gitignore` rules
- Quick exclusion presets:
  - Exclude test files (`__tests__/`, `*.test.*`, `*.spec.*`)
  - Exclude documentation (`docs/`, `*.md` except README)
  - Exclude config files (`*.config.*`, `.*rc`)
- Custom file extension filtering

### 3. Dependency Metadata Extraction
- Parses `package.json` for npm dependencies
- Parses `requirements.txt` for Python packages
- Includes dependency information in output header
- Helps LLMs understand project technology stack

### 4. Chunked Output
- Configurable chunk size (in tokens)
- Automatically splits large repositories into multiple files
- Each chunk is self-contained with proper XML wrapping
- Designed for models with smaller context windows

### 5. Watch Mode
- Polls file system for changes every 5 seconds
- Re-syncs file list when changes detected
- Useful for iterative development with AI copilots

## Use Cases

### For Developers (Vibe Coding)
- **Full codebase review**: "Review this entire codebase for architectural issues"
- **Refactoring planning**: "Plan a migration from Express to Hono"
- **Security audit**: "Find potential security vulnerabilities in this project"
- **Test generation**: "Generate comprehensive tests for all untested functions"
- **Documentation**: "Generate API documentation from these source files"

### For Technical Documentation Teams
- **Gap analysis**: "Find undocumented public APIs in this codebase"
- **Consistency check**: "Verify docs match the current implementation"
- **Migration guides**: "Generate a migration guide from v2 to v3 based on the diff"

### For Authors & Writers
- **Manuscript review**: "Review this entire novel for plot inconsistencies"
- **Consistency check**: "Find character name variations across all chapters"
- **Structural analysis**: "Analyze the pacing across the full manuscript"

## Technical Stack
- React 19 with TypeScript
- Vite 6 build system
- TailwindCSS v4 for styling
- Lucide React for iconography
- Motion library for animations
- File System Access API for local file reading

## Pricing
Free. Open Source. Apache 2.0 License.

## Links
- Website: https://repo2llm.com
- GitHub: https://github.com/getwebface/Repo2LLM
- Docs Audit: https://repo2llm.com/docs-audit
- For Authors: https://repo2llm.com/for-authors
- Sponsor: https://buymeacoffee.com/repo2llm

## Contact
- Twitter: https://twitter.com/getwebface
- GitHub Issues: https://github.com/getwebface/Repo2LLM/issues

## Structured Data Summary (for AI agents)
- Name: Repo2LLM
- Type: SoftwareApplication / DeveloperTool
- Category: Context Engineering, Prompt Engineering, Vibe Coding
- Platform: Browser (Chrome, Edge; fallback for Firefox, Safari)
- Cost: Free
- License: Apache-2.0
- Privacy: Zero-knowledge, client-side only
- Output Format: XML-tagged text files (Anthropic standard)
- Supported LLMs: GPT-4o, GPT-4.5, GPT-5, Claude 3.5, Claude 4, Gemini 1.5, Gemini 2.0