Agent-S

Agent S, developed by Simular AI, is an open-source framework enabling autonomous computer interaction via an Agent-Computer Interface (ACI). By pairing major LLMs with visual grounding models like UI-TARS, it translates natural language into precise mouse and keyboard actions. The latest Agent S3 features a local coding environment for Python/Bash execution and Behavior Best-of-N mechanisms. It demonstrates strong zero-shot generalization across platforms and is the first GUI agent to surpass human-level performance (72.60%) on the OSWorld benchmark.

• First GUI agent to surpass human-level performance (72.60%) on OSWorld
• Autonomous OS interaction driven by a robust Agent-Computer Interface (ACI)
• Deeply integrates visual grounding models for highly precise UI coordinate resolution
• Built-in local coding environment for autonomous Python and Bash script execution