stratapilot

Stratapilot is a system-level, general-purpose copilot agent that automates and executes user tasks across applications, files, and interfaces. Designed for OS-native deployment, Stratapilot enables context-aware, multi-modal task execution through natural language input.

At the core of Stratapilot is the Model Context Protocol (MCP) — an extensible, application-agnostic protocol that allows Stratapilot to interface with third-party software (both legacy and modern) in a structured, deterministic way. MCP defines a standard for model-to-model and model-to-application communication, making Stratapilot interoperable across tools, services, and agents without requiring custom plugins.

⚡️ Quickstart

1. Clone the repository:

   git clone https://github.com/KAIST-KEAI/stratapilot.git
   

2. Set up Python environment (3.10+):

   conda create -n stratapilot_env python=3.10 -y
   conda activate stratapilot_env
   

3. Install dependencies:

   cd stratapilot
   pip install -e .
   

4. Configure your backend LLM API key: Set your LLM provider's API key (e.g., OpenAI) in the .env file and select your preferred model.

5. Run the demo:

   python quick_start.py
   

6. Set up GUI mode (optional):

   Specify the default agent prompt in a configuration file named prompt.conf and place it in the root directory under the cloned project. You can find an example prompt file in prompt.py. After this is done, enable GUI mode and start the program with

   python quick_start.py --gui
   

   The prompt in the configuration file is used to fine-tune the agent's behavior upon startup, and will not be visible in the chat history. You can then chat with the agent like with any LLM chat bot services.

🛠️ Advanced Usage

Stratapilot provides a developer SDK for building:

• Custom MCP Workflows: Compose complex, multi-step tasks that span filesystems, APIs, user interfaces, and third-party apps. Each workflow is defined declaratively and executed deterministically by the agent engine.
• Adapters for Legacy Software: Wrap older desktop or command-line applications with an MCP-compatible interface, enabling structured communication with modern agents without modifying the original software.
• Headless Agents and Background Tasks: Run silent or daemon-mode agents that respond to triggers (e.g., cron jobs, file changes, system events) instead of user queries.
• Plugin-Free Automation: Stratapilot doesn’t rely on per-app plugins or brittle scripting. With MCP bindings, you can interface with existing applications through their CLI, APIs, sockets, or filesystem I/O.

App Integration

Stratapilot supports a variety of integration methods out-of-the-box:

• File-level automation (read/write/transform data files)
• Command-line tool orchestration (wrap tools like ffmpeg, git, curl, etc.)
• API interactions (REST/GraphQL endpoints through typed descriptors)
• Database queries (e.g., SQLite, Postgres via connector wrappers)
• System and window control (e.g., launching apps, recording input/output state)

See sample code under examples for SDK usage demos and integration templates. For more information about tool development using the SDK, refer to Stratapilot's documentation.

🤗 Contributing

Stratapilot welcomes contribution in any form! If you are interested in getting involved, refer to our contribution guidelines.