Maya is an elite, open-source coding-agent CLI optimized for the Google Gemini 3.1 ecosystem and local model providers.
Developed by Ruchir Joshi, Maya was inspired by the work of the OpenClaude project. As students, we often have access to powerful tools like Gemini AI Pro through our university subscriptions, yet many high-end agentic tools remain locked behind proprietary paywalls like Claude Code. Maya was built to bridge this gap—providing a professional-grade, agentic terminal for everyone who wants the power of an elite AI engineer without the proprietary price tag.
Use OpenAI-compatible APIs, Gemini 3.1 (Pro/Flash), GitHub Models, Codex, Ollama, and more, while keeping one unified, terminal-first workflow.
Quick Start | Why Maya | Setup Guides | Providers | Source Build
- Optimized for Students & Developers: Deeply tuned for Gemini 3.1 Pro and Flash, taking full advantage of the 2M token context window available in many academic/developer tiers.
- Elite Agentic Power: Research, plan, and execute multi-file refactors autonomously. Maya doesn't just chat; it works your terminal.
- Free & Open Source: Built so that the engineering power of agentic tools is accessible to everyone, regardless of whether they can afford a Claude subscription.
- Unified Workflow: One CLI for every backend. Switch between Gemini, local Ollama models, or OpenAI with a single command.
- Project Native: Full support for
MAYA.mdand.maya/project standards to keep your agent aligned with your codebase.
# Clone the repo
git clone https://github.com/ruchjoshi-nyu/maya.git
cd maya
# Install and build
bun install
bun run build
# Run locally
./bin/mayaMaya is optimized for Gemini 3.1. Ensure your GEMINI_API_KEY is set:
export GEMINI_API_KEY=your-key-here
./bin/mayaInside Maya, run /init to automatically generate a MAYA.md file for your project.
| Provider | Optimization Level | Notes |
|---|---|---|
| Google Gemini | 🚀 Highest | Optimized for Gemini 3.1 Pro. Uses native tool-calling and huge context windows. |
| OpenAI | ✅ High | Works with GPT-4o, GPT-4o-mini, and custom endpoints. |
| Ollama | ✅ High | Local inference with no API key required. |
| GitHub Models | ✅ High | Interactive onboarding for students with GitHub Copilot access. |
- Autonomous Coding: Bash, file read/write/edit, grep, glob, and specialized engineering agents.
- Deep Research: Maya uses a "Research -> Strategy -> Execution" cycle to handle complex bugs.
- Massive Context: Specifically tuned to handle the large-scale context windows of the Gemini 3.1 series.
- MCP Support: Connect Maya to external tools and data via the Model Context Protocol.
Maya began as a personal project by Ruchir Joshi to make elite AI agent tools accessible to the student community. Inspired by the architectural foundations of OpenClaude, Maya has been rebranded and tuned specifically for the Gemini 3 series of models.
The goal is simple: Any model. Every tool. Zero limits.
Maya is an independent community project. It is a derivative work inspired by the OpenClaude project and the original engine principles of Claude Code. "Maya" is not affiliated with, endorsed by, or sponsored by Anthropic or Google.
See LICENSE.