# Ryu

> Open-source agentic infrastructure for deploying, running, and governing AI agents locally, self-hosted, or in a managed cloud.

Ryu is built by A Major Pte. Ltd. in Singapore and helps teams deploy and manage AI agents by connecting models, tools, workflows, permissions, and usage tracking. It is labeled a Research Preview, with release v0.5.1 listed on GitHub. Core and Gateway are open source and can be self-hosted, and a desktop app is available for download.

## What It Is

Ryu is agentic infrastructure rather than another framework for writing agents. According to the README, it runs the agents and harnesses you already use, such as Claude Code, Codex, OpenClaw, or your own, with the models you choose, on your own machine, on your server, or in Ryu Cloud. Interfaces include a desktop app, web, CLI, mobile, bots, and channels, plus an SDK for defining agents, workflows, tools, and skills in code.

## How the Platform Fits Together

Ryu Core (default port 7980) decides what runs: which agent, session, workflow, or tool. It assembles context such as memory, allowlisted tools, and persona. Ryu Gateway (port 7981) decides what is allowed, routed, measured, and paid for, covering routing, a firewall, budgets, caching, evals, org policy, and audit. In a chat request, Core receives the call, picks the agent, forwards to the Gateway with a Gateway token, and streams the response back to the client. If the Gateway is configured but unreachable, Core falls back to the direct provider path and Gateway policy does not apply to that turn.

## Engines and Models

Ryu runs around existing engines, including OpenAI, Anthropic, Google Gemini, OpenRouter, ACP agents such as Claude Code, Codex, and Pi, and local models. On first run, Core downloads llama.cpp and Gemma 4 and chats locally. Voice, image, and embedding engines run alongside the chat engine.

## Deployment Models

The site describes three options: a self-hosted Ryu Node (Core and Gateway on your own machines, usable without a Ryu account), Ryu Cloud (a managed node hosted in the EU or Singapore), and a Private/Full Ryu Deployment with a Ryu control plane around customer-operated nodes. Core is Apache-2.0 and Gateway is AGPL-3.0; Desktop and some UI packages are source-available under a commercial license. The README notes Ryu is pre-1.0, so interfaces, APIs, and on-disk formats may change between releases.

## Governance and Enterprise Controls

The enterprise page lists OIDC and SAML single sign-on, SCIM directory sync, roles, budgets and rate limits, IP allowlists, audit log export, signed audit streaming, and zero data retention for prompts routed through Ryu. SSO, SCIM, and audit export require an Enterprise entitlement or license and are not part of the self-hosted node.

## Features
- Core runtime for agents, sessions, workflows, and tools
- Gateway with routing, firewall, budgets, caching, and audit
- Bring your own agent engine (Claude Code, Codex, OpenClaw, ACP agents)
- Local models via llama.cpp and Gemma 4
- Hosted model providers: OpenAI, Anthropic, Google Gemini, OpenRouter
- MCP tools, skills, plugins, and Marketplace
- SDK and CLI for defining agents, workflows, tools, and skills
- Desktop app
- Telegram and Discord bot channels
- Cross-device sync (opt-in)
- Self-hosted, managed cloud, or private deployment
- Enterprise SSO (OIDC/SAML), SCIM, RBAC, and audit export

## Integrations
Claude Code, Codex, OpenClaw, OpenAI, Anthropic, Google Gemini, OpenRouter, llama.cpp, MCP, Telegram, Discord

## Platforms
MACOS, LINUX, WINDOWS, WEB, CLI, API, DEVELOPER_SDK

## Pricing
Freemium — Free tier available with paid upgrades

## Version
v0.5.1

## Links
- Website: https://ryuhq.com
- Documentation: https://docs.ryuhq.com/
- Repository: https://github.com/amajorai/ryu
- EveryDev.ai: https://www.everydev.ai/tools/ryu
