Simular

Simular builds AI agents that carry out tasks in desktop and web applications. Its Agent S framework takes natural-language instructions, reads the screen, and clicks, types and scrolls through ordinary software interfaces. The open-source framework serves researchers and developers exploring computer automation, with support for macOS, Windows and Linux.

Sai, Simular’s hosted computer-use agent, turns repeating jobs into scheduled workflows on cloud computers. Users describe a routine and choose when it should run. Developers can also submit plain-English tasks through the Simular API’s HTTP endpoint: Sai opens applications, performs the work and returns the result. This provides access to computer-use agents without requiring developers to operate virtual machines, display drivers or an agent harness themselves.

Simular’s approach combines adaptable visual agents with reusable automation code. Successful steps can be saved as workflows and replayed; when a fixed script encounters a change, a model can explore again and help recover. Checking the environment during execution supports that balance between repeatable automation and adaptation. The technical challenge extends beyond completing a task once to completing it reliably as instructions and application conditions vary.