Orbit framework gives developers structured Python control over AI agents that automate computer tasks without relying on screenshots.

Orbit bridges the gap between opaque AI frameworks and raw tool calls by allowing natural language to control screen actions while Python manages workflow logic

Each step in the agent workflow has its own model, budget, and typed output, enabling cost optimization by mixing cheap and expensive models as needed

Built on OS accessibility trees rather than screenshots, allowing extraction of structured data into Pydantic models and mid-task course correction when agents struggle

Shares context across the entire session while maintaining individual step isolation, providing both flexibility and control for computer use automation tasks