Build and run AI agents without starting from scratch. Extend capability with plugins, or turn agents into apps.
Ryu is a platform for building, running, and governing AI agents. It provides the layer around an agent: tools, memory, workflows, model access, permissions, and delivery through apps, plugins, APIs, and clients.
Most teams rebuild these parts for every agent. Ryu gives them one shared layer. Keep the agent or model you already use, then add the capabilities you need.
Ryu is for teams that want to move from an agent demo to repeatable work without rebuilding the stack for every job.
- Start with what you have. Connect Claude Code, Codex, Pi, OpenClaw, or another ACP or OpenAI-compatible agent. Use local models, hosted models, or both.
- Add the missing layer. Use agents, workflows, teams, plugins, MCP servers, skills, apps, and a marketplace to package repeatable work.
- Keep control. Route model calls and enforce permissions, budgets, approvals, firewall rules, durable state, and audit history.
- Swap parts as you grow. Replace models, engines, memory, retrieval, sandboxes, and integrations without rebuilding the product around them.
- Run it where work happens. Use the same system through the CLI, Desktop, Web, bots, SDKs, or an API, with self-hosted Core and Gateway or managed options.
This public repository includes the open-core runtime, Gateway, CLI/TUI clients, SDKs, capability crates, source-available Desktop, Island, shared UI packages, and build/deploy files for the self-hosted stack. The Web, server, mobile, browser-extension, identity, and billing surfaces live outside this repository; the full product documentation is at docs.ryuhq.com.
Ryu sits at the control-plane and runtime layer around an agent. The matrices separate agent frameworks and local runtimes from managed agent products and cloud platforms. They describe documented product surfaces, not model quality, latency, or security certification.
Legend: ✅ first-class documented capability · 🟡 available through composition, configuration, or a limited/preview surface · ❌ not the product's primary documented surface.
The hosted eve/Vercel path is usage-metered: Vercel charges for Function active CPU,
provisioned memory time, and invocations. Ryu's self-hosted path runs on infrastructure you
control. See Vercel Functions pricing.
| Capability | Ryu | Mastra | LangChain / LangGraph | eve | OpenClaw | Hermes Agent | Omnigent |
|---|---|---|---|---|---|---|---|
| Use an existing agent or harness | ✅ | ❌ | ❌ | ❌ | ❌ | 🟡 | ✅ |
| BYO models and providers | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Local or self-hosted runtime | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Managed hosting | ✅ | 🟡 | ✅ | 🟡 | ❌ | 🟡 | 🟡 |
| Multi-agent coordination | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Durable workflows and triggers | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | 🟡 |
| Tools and MCP | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Memory and retrieval | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | 🟡 |
| Routing, budgets, approvals, and audit | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | ✅ |
| Sandboxed execution | ✅ | 🟡 | 🟡 | ✅ | 🟡 | 🟡 | ✅ |
| Background schedules | ✅ | ✅ | 🟡 | ✅ | ✅ | ✅ | 🟡 |
| CLI, app, and channel delivery | ✅ | 🟡 | 🟡 | ✅ | ✅ | ✅ | ✅ |
| Skills, plugins, and integrations | ✅ | 🟡 | ✅ | ✅ | ✅ | ✅ | 🟡 |
| Capability | Ryu | Claude Managed Agents | Notion Custom Agents | Grok Bot | Hyperagent | Vercel AI Cloud | ChatGPT Workspace Agents | Microsoft Foundry | Vertex AI Agent Engine |
|---|---|---|---|---|---|---|---|---|---|
| Use an existing agent or harness | ✅ | 🟡 | ❌ | ❌ | 🟡 | 🟡 | ❌ | ✅ | ✅ |
| BYO models and providers | ✅ | ❌ | 🟡 | ❌ | 🟡 | ✅ | 🟡 | 🟡 | 🟡 |
| Local or self-hosted runtime | ✅ | 🟡 | ❌ | ❌ | ❌ | ❌ | ❌ | 🟡 | 🟡 |
| Managed hosting | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Multi-agent coordination | ✅ | ✅ | 🟡 | 🟡 | ✅ | 🟡 | 🟡 | ✅ | ✅ |
| Tools and MCP | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | ✅ | ✅ | 🟡 |
| Memory and retrieval | ✅ | ✅ | ✅ | ❌ | ✅ | 🟡 | 🟡 | 🟡 | ✅ |
| Routing, budgets, approvals, and audit | ✅ | 🟡 | ✅ | 🟡 | ✅ | 🟡 | 🟡 | ✅ | 🟡 |
| Sandboxed execution | ✅ | ✅ | ❌ | 🟡 | 🟡 | ✅ | 🟡 | 🟡 | 🟡 |
| Background schedules and triggers | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | 🟡 | 🟡 |
| CLI, app, and channel delivery | ✅ | 🟡 | ✅ | ✅ | ✅ | 🟡 | ✅ | 🟡 | 🟡 |
| Skills, plugins, and integrations | ✅ | ✅ | ✅ | 🟡 | ✅ | 🟡 | ✅ | ✅ | 🟡 |
The self-hosted path runs Core and Gateway on infrastructure you control. It does not need a Ryu account or control-plane service.
macOS and Linux:
curl -fsSL https://raw.githubusercontent.com/amajorai/ryu/main/install.sh | sh
ryu-cliWindows PowerShell:
irm https://raw.githubusercontent.com/amajorai/ryu/main/install.ps1 | iex
ryu-cliThe installer starts Core and the Gateway with the bundled local defaults. Read the self-hosting guide for providers, deployment, and configuration.
The open-core units are Apache-2.0, except the Gateway, which is AGPL-3.0. Desktop, Island,
and shared UI packages are source-available under
LICENSE-COMMERCIAL.md; they are not open source. See
LICENSING.md and TRADEMARK.md for the full boundary.
Ryu is pre-1.0. Interfaces, APIs, and on-disk formats may change between releases.
Contributions to the open-source units and source-available layer are welcome. Open a pull request in this repository and start with the contribution guide. Accepted changes are carried back into the Ryu monorepo by maintainers and included in a later sync. Report security issues through SECURITY.md.
