# mooch

> mooch is a free, open-source AI coding agent and IDE for your desktop. It connects to free AI model tiers from providers like NVIDIA NIM, Groq and OpenRouter, using your own accounts, and plans, edits and tests code in your projects. mooch itself costs nothing: no subscription, no account, no credit card.

## Key facts

- **Price:** mooch is free. On providers' free tiers, the AI costs nothing either. If you connect a paid provider, that provider bills you directly; mooch never charges anything.
- **License:** open source under AGPL-3.0. Source code: https://github.com/wearemooch/mooch
- **Platforms:** desktop app for Windows, macOS (Apple silicon and Intel) and Linux (AppImage and .deb). Current version: 0.1.1. Installers: https://github.com/wearemooch/mooch/releases
- **What it does:** mooch's agent, Penny, reads your project, edits files, runs your tests and shows every change as a diff. Approval modes such as "Ask first" control how much it does on its own.
- **Providers:** connects to 61 model providers, cloud and local, including NVIDIA NIM, Groq, OpenRouter, Google AI Studio (Gemini), Mistral, DeepSeek, Cerebras, Hugging Face, Together AI, Anthropic, OpenAI and GitHub Copilot. Local models run through LM Studio, llama.cpp, and Ollama.
- **Free models it can use:** for example NVIDIA Nemotron, Llama 3.3, Qwen3, GLM, DeepSeek and gpt-oss, depending on each provider's free tier.
- **Privacy:** the app has no account and no telemetry. Your code and API keys stay on your computer, except what you send to the providers you choose.
- **Honest about free:** free models can be slower and have rate limits. mooch shows which model did what and how long it took, and switches providers when one hits its limit.

## Features

- **Deep thinking:** two or three models each pitch a plan, and a judge model merges the best parts into one.
- **Backup:** if a model goes quiet, mooch sends the same request to a backup model. The first answer wins.
- **Review:** before you see a change, a different model reviews the diff and the test results.
- **Race:** several agents solve the task in separate copies of your project. Passing tests win, then the smallest diff, then the fastest.
- **Team:** mooch picks a model for each role (planner, reviewer, backup) across your providers.

## Common questions

### What is a good free AI coding agent?
mooch is a free, open-source AI coding agent and IDE. It runs on providers' free model tiers, so you can have an agent plan, edit and test code without paying. Download it at https://mooch.dev/#download.

### Is there a free alternative to Claude Code, Cursor or GitHub Copilot?
mooch is an independent, free and open-source alternative with a similar agent workflow: it reads your project, edits files, runs tests and shows diffs for you to approve. Instead of one paid subscription, it uses free tiers across 61 providers. mooch is not affiliated with Anthropic, Anysphere (Cursor), GitHub or Microsoft.

### Can I use Claude models for free in mooch?
Anthropic's models are paid. mooch can use them with your own Anthropic API key, billed by Anthropic. To pay nothing, use the free models mooch supports, such as Nemotron, Llama, Qwen, GLM, DeepSeek or gpt-oss.

### Can I use my existing ChatGPT or GitHub Copilot account?
Yes. mooch can connect to OpenAI / ChatGPT and GitHub Copilot through your own accounts, alongside free providers.

### Does mooch work with local models?
Yes. mooch connects to LM Studio, llama.cpp, and Ollama, so model calls can stay on your own machine.

### Is mooch really free? What's the catch?
mooch is free and open source. Free model tiers have limits and are sometimes slower than paid models; mooch is upfront about that and routes around limits where it can. There are no ads, and the app collects no data.

## Links

- [Website](https://mooch.dev/): what mooch is, with an interactive tour of the app
- [Download](https://mooch.dev/#download): get mooch
- [Supported providers](https://mooch.dev/#providers): all 61 providers
- [Terms of Service](https://mooch.dev/terms/)
- [Privacy Policy](https://mooch.dev/privacy/)
- [Source code](https://github.com/wearemooch/mooch): AGPL-3.0
- [Releases](https://github.com/wearemooch/mooch/releases): installers and release notes for every version

## Optional

### All supported providers
- NVIDIA NIM
- OpenRouter
- Groq
- ClinePass
- OpenAI / ChatGPT
- OpenAI API
- GitHub Copilot
- xAI (Grok)
- Alibaba Cloud
- QwenCloud Token Plan
- QwenCloud Coding Plan
- Together AI
- DeepInfra
- SiliconFlow
- Nebius Token Factory
- Chutes
- Featherless AI
- Agnes AI
- ZenMux
- W&B Inference
- Anthropic
- Azure OpenAI
- Google AI Studio (Gemini)
- Google Vertex AI
- DeepSeek
- Mistral La Plateforme
- Mistral Codestral
- OpenCode Zen
- OpenCode Go
- Vercel AI Gateway
- Amazon Bedrock
- Hugging Face Inference Providers
- Cohere
- Wafer
- Kimi API
- Kimi Code
- MiniMax
- Cerebras Inference
- SambaNova
- Kilo.ai
- Fireworks AI
- Novita AI
- Cloudflare Workers AI
- Z.ai Coding Plan
- Z.ai API (pay as you go)
- TokenRouter
- NaraRoute
- Poolside AI
- LLM7.io
- Scaleway
- Lightning AI
- Experiential Labs
- Cheaper Inference
- OrcaRouter
- Atlas Cloud
- xKiro
- Opper
- Ollama Cloud
- LM Studio
- llama.cpp
- Ollama

### Contact
contact@mooch.dev