Privacy-first AI coding, no cloud lock-in.
JarvisCodex exists to give developers the power of AI coding assistants without surrendering their code, their data, or their choice of model. We believe the best AI tools should run locally, respect your privacy, and let you bring your own API keys.
JarvisCodex is a desktop IDE with an embedded multi-agent runtime. It connects to any LLM provider — OpenAI, Anthropic, Google, Mistral, or local models via Ollama — through your own API keys (BYOK). Your code never touches our servers.
Built with Electron, React, TypeScript, and Vite. The agent runtime is a standalone TypeScript package that handles planning, tool routing, staged execution, and result compression. We use SQLite for local persistence and Stripe for billing.
We develop in the open on GitHub. Feature requests, bug reports, and contributions are welcome.
Have questions or want to partner with us?