Your AI agents, on your terms.
OpenCinna is open-source software for people who want real AI agents without becoming infrastructure engineers. Spin up your own platform on your own hardware, build agents of any complexity in a clean web UI, and reach them from anywhere through a native desktop app. No cloud lock-in, no YAML rituals, no plumbing — your agents, your data, your machines.
Most agent frameworks ask you to be three things at once: a developer, a DevOps engineer, and a prompt scientist. We don't think that's how this should feel.
OpenCinna is built around one idea: the smartest things in AI should be available in the easiest possible way. That means:
- Self-hosted by default — run the whole platform on your own hardware, your own VPS, or your own laptop. Your agents, your data, your control.
- Build in the platform, talk from the desktop — agents are built in the platform's web UI (or with
cinna-cliif you prefer your own editor). The desktop app is for using them: chatting, sending tasks, getting work done. - Agents of any complexity, without the complexity — from a one-prompt assistant to a multi-agent team with scheduled jobs, email triggers, and external tool access. The hard parts are handled for you.
- Use them right away — agents are runnable the moment you save them. Chat with them, schedule them, expose them over MCP, or hand them tasks via email.
- Built for sharing — every agent can be shared with other users on your instance. One person builds the "AI employees"; the whole team uses them. A single technically-minded person can equip an entire small company with custom assistants — HR, finance, support, ops — without anyone else touching a prompt or a config file.
- Open and inspectable — every layer is open source. No black boxes, no usage caps, no surprise bills.
- Build an agent in minutes through a guided wizard in the web UI, or develop locally in your own editor with
cinna-cli - Give agents real capabilities: shell access in an isolated Docker workspace, files, credentials, knowledge bases, scheduled triggers, webhooks, email automation
- Share any agent with teammates, or publish it as a bundle that other users on your instance can install with one click — like apps in an app store, but yours and your team's
- Reach your agents from the desktop app, the web UI, email, MCP-compatible clients (Claude Desktop, Cursor, etc.), or other agents via the A2A protocol
- Compose agentic teams with handover, delegation, and visual org-chart routing
- Build per-agent dashboards and lightweight web apps served straight from the agent's workspace
All of this without writing deployment scripts, managing Kubernetes, or wiring queues.
| Repo | What it is |
|---|---|
| cinna-core | The platform — FastAPI backend, React + TypeScript frontend, Docker-isolated agent runtimes. Build, run, and manage agents from the web UI. |
| cinna-desktop | The desktop app — native client for talking to your agents from any machine. Sign in to your instance, chat, send tasks, get answers. Communication only — building lives in the platform. |
| cinna-cli | Local development CLI — bidirectional workspace sync and remote exec for agent developers who want to build with their own editor instead of (or alongside) the web UI. |
OpenCinna separates three things so each can stay simple:
- Agent — what it does (prompts, credentials, integrations, SDK config)
- Environment — where it runs (an isolated Docker container with the agent's workspace and tools)
- Session — how you talk to it (a persistent chat thread; one environment can host many)
Underneath: PostgreSQL for state, SQLModel for the data layer, FastAPI for the API, React + TanStack on the frontend, and pluggable LLM backends (Anthropic, OpenAI, Google, MiniMax, OpenAI-compatible). Agents run on top of the Claude Code or OpenCode SDK, with MCP and A2A as first-class integration protocols.
Pick your entry point:
- I want to use agents → install the desktop app, point it at any OpenCinna instance, sign in, install bundles, start chatting. The web UI works too if you'd rather stay in the browser.
- I want to host my own instance → clone
cinna-core, bring it up with Docker Compose, and you have a full platform on your hardware. - I want to build agents → spin up an instance, log into its web UI, get promoted to the developer role, and use the in-app builder. Prefer your own editor? Pair it with
cinna-clifor live workspace sync.
Each repo has its own setup guide.
OpenCinna is under active development. The core platform is functional — bundles, multi-SDK support, MCP/A2A integration, agentic teams, and the desktop client are live — but expect rapid iteration, breaking changes, and the occasional sharp edge. As with any open-source software, use at your own risk: review the code, test in your own environment, and keep your data backed up. No warranty is provided.
Open source. See the LICENSE file in each repository for specifics.
OpenCinna — your AI agents, on your terms.