Code with agents that think, act, and ship.
Rayu is a multi-provider Agentic AI coding agent. Bring your own API key, connect any model, and ship faster with autonomous agents that write, test, and deploy your code.
Engineered for hyper-growth development
All the tools you need to build, test, and ship from a single command prompt.
Multi-Provider
Connect any LLM with the simple /connect command. Switch providers on the fly.
MCP Servers
Unlock 600+ tools through Model Context Protocol — databases, browsers, file systems, and more.
Autonomous Agents
Deploy agents that read, think, plan, write, test, and correct their own errors automatically.
Terminal Native
Built for keyboard warriors. Full shell access, git integration, local commands.
Persistent Memory
Retains context across sessions — project specifics, past bugs, and conventions remembered automatically.
Open Source
Self-hostable, customizable, BYOK forever free. Run entirely locally with open-weights models.
Why Rayu outperforms the rest
Built from the ground up to outperform every other CLI agent on the market. Faster, private, and provider-agnostic.
| Feature | Rayu | OpenCode | KiloCode | Claude Code | Codex |
|---|---|---|---|---|---|
| Response Speed | <500ms | ~3–5s | ~5–10s | ~5–8s | ~3–6s |
| P2P Collaboration | Native | ||||
| Privacy / No Training Data | Zero retention | Cloud-dependent | Cloud-dependent | Trains on your code | Trains on your code |
| Multi-Provider BYOK | 20+ providers | Limited | Single | Anthropic only | OpenAI only |
| Agent Swarms | Built-in & parallel | ||||
| Offline / Local Models | Private local support |
One CLI. Every provider you use.
Bring your own key or use the Rayu-hosted gateway. Switch models mid-session with /model — no rewrites, no lock-in.
What our clients say about us
Real feedback from engineers who ran Rayu CLI on real work.
It's cool that Cambodia gets to own a tool like this, instead of just reaching for Claude Code every time.
What sets it apart from OpenCode and the rest of the open-source field is per-agent model assignment — Gemini on frontend, Claude Sonnet on backend, Claude Opus running the orchestrator.
Genuinely fun to use, and faster than Claude Code or OpenCode at getting to a good result.
It ships image and video generation built in, so a frontend dev can stay in one terminal for the whole workflow.