MIRA is an open-source AI agent tooling for long-running research work.

MIRA helps you go from idea -> experiments -> report in one workflow. It is built for project-based research, not one-off chat, so every run leaves behind traceable artifacts such as workspace files, experiment outputs, and task_plan.json.

If you are new to MIRA, start with the desktop bundle.

1. Download the latest MIRA Desktop Bundle from mira-ui releases.
2. Open the app and keep Local mode selected.
3. In Settings -> Local Engine, fill in your provider, model, and API key.
4. Create your first project and let MIRA plan work, run experiments, and export results.

For most users, this is the fastest way to get started.

• Recommended for most users: MIRA Desktop downloads
• Docs and product overview: mira-intelligence.github.io
• CLI / engine binaries and Python package: mira releases

• Choose MIRA-bundle-* if you want the easiest local setup. It includes the desktop app plus a bundled local engine.
• Choose MIRA-standalone-* if you already run mira separately or want the UI to connect to a remote deployment.
• Choose mira-engine-* or the Python package from mira releases if you only need the engine / CLI side.

• Built for long-running research tasks: organize work around workspace, project, experiment, and task plan instead of a single chat thread.
• 20+ built-in model providers: works with OpenAI, Anthropic, OpenRouter, DeepSeek, Ollama, Azure, vLLM, and more.
• CLI / Web / Desktop: use the same system in the terminal, browser, or packaged desktop app.
• 13 channel integrations: connect MIRA to tools like Feishu, Slack, Telegram, email, Matrix, and others.
• Guardrail auto-repair: MIRA can repair incomplete experiment outputs before handing work back to you.
• Local-first and self-hostable: data stays under your control and can run on one machine or in your own deployment.

Use the desktop bundle on macOS or Windows, configure your model provider, and start projects locally.

Install mira on a remote machine, then connect to it from MiraUI standalone or the web UI.

• Concepts
• Provider and runtime configuration
• Channel integrations
• Desktop UI guide
• Deployment options

If you only want to try MIRA as quickly as possible, download the latest MIRA-bundle-* package and start there.

If you need remote deployment, team setup, or server-side control, install mira separately and connect the UI to that deployment.