Agent S (12.6k stars, Apache-2.0, Agent S3 release) from Simular is a computer-use framework: give the agent_s CLI a task and it takes screenshots and clicks, types and scrolls in ordinary desktop and web apps on Windows, macOS or Linux. It needs a main model API (OpenAI, Anthropic, Gemini, OpenRouter, Azure or vLLM) plus a grounding model such as UI-TARS-1.5-7B on a Hugging Face endpoint or your own server. Single monitor only. A research framework with no GUI app; recent commits are mostly README updates.
Cite this pageRepoQuarry. “Agent S.” Checked 16 Oct 2026. https://repoquarry.dev/projects/agent-s
How much of Manus does it replace?
1 of 6 jobs. Drives your own desktop and browser; no sandbox, research reports, slides, schedules or connectors.
Partial:Autonomous web browsingPartial replacement
No:Deep research reportsNot covered
No:File and code tasks in a sandboxNot covered
No:Slides generated from a promptNot covered
No:Scheduled agent tasksNot covered
No:Connectors to your appsNot covered
For deep research reports, file and code tasks in a sandbox, slides generated from a prompt, scheduled agent tasks, connectors to your apps, see other Manus alternatives. You may also need to build an interface, an API server and hosting yourself.
Facts, with sources
Every answer links to where we found it. We read the documentation; we did not install it.
Runs on my machine?Confirmed:Yes, runs on your computer and controls its desktop; models can be remote APIsIt takes a natural-language task, looks at the screen, and completes the task by clicking, typing, and scrolling in ordinary desktop and web applicationsgithub.com · Agent-S#-introduction ↗today
Model weights licenceConfirmed:Recommended grounding model UI-TARS-1.5-7B is Apache-2.0; the main model is whichever API you bringhuggingface.co · UI-TARS-1.5-7B ↗today
Commercial useConfirmed:Yes, Apache-2.0 code and recommended grounding weights; main model terms depend on your providergithub.com · LICENSE ↗today
Paid services needed?Partial:Needs a main model API and a grounding model endpoint; both can be self-hosted with vLLM, otherwise paid APIs or Hugging Face endpointsWe support Azure OpenAI, Anthropic, Gemini, Open Router, and vLLM inference.github.com · Agent-S#supported-models ↗today
HardwareConfirmed:A single-monitor desktop; self-hosting the 7B grounding model needs a GPU (size not documented)Our agent is designed for single monitor screensgithub.com · Agent-S#prerequisites ↗today
Setup effortConfirmed:Medium; pip install gui-agents and tesseract, set API keys, and stand up a grounding model endpointpip install gui-agentsgithub.com · Agent-S#installation ↗today
Your choice on cookies
We use essential storage to keep the site secure and remember this choice. With your permission we'd also count visits to see which pages help people, and remember your filters. We never use advertising or cross-site tracking cookies, and sponsors never get your data. Cookie policy