Agent S

Agent S is a computer-use AI agent framework for developers and researchers building desktop automation. It takes natural-language tasks and operates desktop and web applications through their screen interfaces, without requiring an API integration for each app.

• Desktop control: The agent reads the screen and clicks, types and scrolls to carry out tasks in ordinary applications.
• Model choice: Use OpenAI, Azure OpenAI, Anthropic, Gemini or OpenRouter, with vLLM inference also supported for open-weight models.
• Screen targeting: A required grounding model identifies where the agent should act on screen; UI-TARS-1.5-7B is the recommended choice.
• Local code execution: An optional coding environment lets the agent run Python and Bash for spreadsheet processing, file operations, document editing and other tasks suited to code.

The framework runs on macOS, Windows and Linux and is open source under Apache 2.0. It's designed for a single monitor. Desktop actions and optional code execution run on your computer, with code using your user permissions. Cloud model providers process model requests remotely and require an internet connection; the recommended grounding model can be hosted on Hugging Face Inference Endpoints or another provider.

Simular also offers Sai as a separate product that operates on devices and cloud computers.