{"slug":"agent-s","name":"Agent S","vendor":"Simular","tagline":"Open computer-use agent framework from Simular, first to pass human-level OSWorld in its S3 release","url":"https://agentboards.org/agent/agent-s","website":"https://www.simular.ai","docs":"https://github.com/simular-ai/Agent-S","repo":"https://github.com/simular-ai/Agent-S","category":"framework","execution":["local"],"license":"Apache-2.0","open_source":true,"pricing":{"model":"byok","summary":"Free and open source under Apache-2.0; you supply your own model API keys","from_usd":null,"free_tier":true,"byok":true},"models":{"backbone":["any"],"bring_your_own_model":true,"local_models":false},"protocols":{"mcp_client":false,"mcp_server":false},"capabilities":{"terminal_exec":true,"browser":true,"multi_file_edit":false,"git_ops":false,"docker_sandbox":false,"multi_agent":true,"headless_ci":false},"context_window":null,"platforms":["macos","linux","windows"],"install":[{"label":"pip","command":"pip install gui-agents"}],"benchmarks":[{"name":"OSWorld","score":"72.60%","source_url":"https://arxiv.org/abs/2510.02250","note":"Agent S3, reported by the authors as first above the 72% human baseline"}],"tags":["computer-use","gui-agent","research","osworld"],"github_stars":12494,"metrics":{"stars":12494,"pushed_at":"2026-09-05T22:02:40Z","archived":false,"open_issues":53,"latest_version":"v0.3.2","latest_version_at":"2025-12-16T17:38:14Z","latest_version_url":"https://github.com/simular-ai/Agent-S/releases/tag/v0.3.2","version_source":"github","checked_at":"2026-10-02T05:04:03.166Z"},"sources":{"overview":"https://github.com/simular-ai/Agent-S","benchmarks":"https://arxiv.org/abs/2510.02250","vendor":"https://www.simular.ai"},"verified_at":null,"adoption":{"score":5.4,"signals":{"pypi_weekly":418,"github_stars":12328,"hn_mentions_1y":14},"measured_at":"2026-09-21"},"scores":{"panel":6.4,"adoption":5.4,"spread":2.5,"facts":0,"board":5.3,"dimensions":{"reliability":6,"usefulness":6.5,"cost":6.5,"longevity":6.7},"per_persona":{"amigo":6.8,"critico":6.3,"profesor":6.8,"inversora":6.3,"jefa":5,"hacker":7.5},"community":null},"rank":null,"category_rank":86,"panel_reviews":[{"persona":"juez","persona_name":"El Juez","ai_generated":true,"verdict":"El Hacker and La Jefa agree the price is zero and disagree on what it costs: he installs a pip package, she counts a machine per user and a hosted visual endpoint.","scores":null,"evidence":[]},{"persona":"amigo","persona_name":"El Amigo","ai_generated":true,"verdict":"Pick Agent S if you need something that drives real desktop applications; pick Browser Use when the work happens inside a browser and nothing else.","scores":{"reliability":6,"usefulness":7,"cost":7,"longevity":7},"evidence":["https://github.com/simular-ai/Agent-S"]},{"persona":"critico","persona_name":"El Crítico","ai_generated":true,"verdict":"The prerequisites say plainly that the agent runs Python code to control your computer, and the board records no container sandbox, so the blast radius is your desktop.","scores":{"reliability":5,"usefulness":6,"cost":7,"longevity":7},"evidence":["https://github.com/simular-ai/Agent-S"]},{"persona":"profesor","persona_name":"El Profesor","ai_generated":true,"verdict":"The 72.60% headline is the agent plus a Behavior Best-of-N wrapper; the agent alone reaches 66% in the 100-step setting against a prior best of 63.4%.","scores":{"reliability":7,"usefulness":7,"cost":6,"longevity":7},"evidence":["https://arxiv.org/abs/2510.02250","https://github.com/simular-ai/Agent-S"]},{"persona":"inversora","persona_name":"La Inversora","ai_generated":true,"verdict":"Simular publishes the papers and sells the hosted version, so the open framework is a recruiting document and a demo funnel for a cloud product.","scores":{"reliability":7,"usefulness":6,"cost":6,"longevity":6},"evidence":["https://www.simular.ai","https://github.com/simular-ai/Agent-S"]},{"persona":"jefa","persona_name":"La Jefa","ai_generated":true,"verdict":"Every user needs a dedicated single-monitor machine and a separate hosted endpoint for the visual component, so the free framework arrives with two infrastructure line items.","scores":{"reliability":4,"usefulness":5,"cost":5,"longevity":6},"evidence":["https://github.com/simular-ai/Agent-S"]},{"persona":"hacker","persona_name":"El Hacker","ai_generated":true,"verdict":"Apache-2.0, pip install gui-agents, keys come from environment variables, and vLLM is a supported inference path so the whole stack can run on hardware I own.","scores":{"reliability":7,"usefulness":8,"cost":8,"longevity":7},"evidence":["https://github.com/simular-ai/Agent-S"]},{"persona":"comediante","persona_name":"El Comediante","ai_generated":true,"verdict":"Surpassed the human baseline on OSWorld, then asked you to install tesseract with Homebrew first.","scores":null,"evidence":[]}]}