microsoft/fara

Fara is a family of frontier computer use agent models designed to automate tasks by interacting with computer interfaces like a human would.

The tool addresses the problem of automating complex workflows that require visual understanding and interaction with graphical user interfaces. Rather than relying on APIs or structured data, Fara agents perceive the screen, reason about what they see, and execute actions such as clicking, typing, and scrolling. This approach enables automation of tasks across diverse applications without requiring integration with each system's internals, making it applicable to workflows that would otherwise be difficult or impossible to automate programmatically.

Fara suits organizations looking to automate repetitive computer-based tasks where traditional automation approaches fall short due to reliance on visual interfaces or unstructured interactions. It is particularly valuable for scenarios involving web browsers, desktop applications, or any graphical environment where the agent must navigate and make decisions based on what appears on screen. Teams considering adoption should evaluate whether their target workflows involve primarily visual navigation and decision-making, as this is where the agent's strengths lie. The project is positioned as a research effort from Microsoft, so adopters should be prepared for an evolving tool that may require adaptation as the models develop.

The project maintains active development with regular updates to the model family and supporting infrastructure. The codebase is structured to support experimentation with different agent configurations and capabilities. Documentation and examples are provided to guide implementation, though the research-oriented nature means users may need to engage with technical details beyond typical library usage. The project accepts community contributions and maintains responsiveness to issues and feature requests.