Rocky is an open source desktop agent, built with pure Python on top of Flut and the OpenAI Agents SDK.

Rocky has no intention of competing with business-oriented products. Instead, it aims to provide a clean, easy-to-follow base for everyone, whether you want to learn how to build an agent or fork it as a starting point for your own project. Rocky believes anyone can build an agent, and the agent community deserves more choices, rather than being locked into any platform.

• Pure Python: the entire app, UI included, is a plain Python process. No complex dependencies or installation steps.
• Polished Cross-Platform Desktop: powered by Flut, Rocky delivers a Flutter-backed desktop experience on Windows, macOS, and Linux.
• Reusable Design and Code: clean separation between Flutter widgets, agentic logic, and tools. Each can be reused independently.

• Shell Environments: Connect to various shell environments to complete user tasks or control them on behalf of the user. Explorer can be used to browse and manage the shell environments.
• Web Search: Search the web, open pages, and find text within fetched pages using ddgs package.
• Skills: Support standard Claude/Codex skills from local skill folders or .zip files.
• MCP: Support standard STDIO and HTTP MCP servers.
• More are coming soon.

Rocky requires Python 3.10+.

1. Install Python

• Windows: install from python.org, or simply type python in a command prompt and follow the prompts.
• macOS / Linux: Python 3 is usually preinstalled. Verify with python3 --version, or install via Homebrew or your package manager if missing. You may need to use python3/pip3 instead of python/pip.

2. Install dependencies

pip install -r ./requirements.txt

3. Run the app

python app.py

On first launch, open the in-app settings to configure a model provider, then start your agentic journey.

Rocky is built on the OpenAI Agents SDK, which is largely provider-agnostic. In principle, Rocky can work with any model provider the SDK supports, directly or indirectly.

Rocky also integrates LiteRT-LM for in-process local models, letting you run models like Gemma 4 on-device without a separate server.

Rocky can connect to a Docker container, a WSL distribution, a remote machine and more. Add an environment in Settings, choose it from the chat header, and Rocky can use it when the selected model supports tools.

Supported environment types:

• Local: run command arrays in the chat workspace on the host OS.
• Docker: run commands inside a local Docker container.
• SSH: run commands on a remote host over SSH.
• WSL: run commands inside a local WSL distribution.
• Docker in WSL: run Docker commands through a WSL distribution.
• Docker over SSH: run Docker commands on a remote host over SSH.

MIT License.