# simular-ai/Agent-S

Agent S: an open agentic framework that uses computers like a human

Repository: https://github.com/simular-ai/Agent-S
Canonical: https://ross.abutalabs.com/products/agent-s
Homepage: https://www.simular.ai
Language: Python
License: Apache-2.0
License Family: permissive
Topics: agent-computer-interface, ai-agents, computer-automation, gui-agents, memory, mllm, planning, retrieval-augmented-generation, in-context-reinforcement-learning, computer-use, grounding, computer-use-agent, cua
Last push: 2026-08-01T03:31:09+00:00

## Health v2 (maintenance only)
Score: 70/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 95, release rhythm 49, longevity 49
- inputs: {"age_days": 693, "days_push": 32, "days_rel": 260, "gap_med": 31, "n_releases_24m": 12}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 12193, forks 1424 (observed 2026-08-28T04:10:51.995862+00:00)

## What it is
Agent S is an open-source agentic framework that uses multimodal LLMs to operate computers like a human, controlling GUIs via clicking, typing, and screen understanding. It includes an Agent-Computer Interface, planning, memory, and self-improving in-context learning, and has surpassed human performance on the OSWorld benchmark.

## Use cases
- automate desktop tasks by controlling the mouse and keyboard with an AI agent
- build a computer-use agent that operates GUI applications
- run an AI agent that completes tasks on Windows, macOS, or Linux
- benchmark GUI agents on OSWorld
- automate repetitive clicking and typing workflows with LLMs
- build agents that see the screen and take actions like a human

## When to choose
- you need an open-source computer-use agent that operates GUIs across Windows, macOS, and Linux
- you want a research-backed framework with state-of-the-art OSWorld results
- you want planning, memory, and self-improvement built into a GUI agent

## When to avoid
- you only need browser automation rather than full desktop control
- you need a fully deterministic RPA pipeline with predefined actions
- you cannot run or pay for the multimodal LLM backends the agent requires

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, computer-vision, rag, machine-learning, gui
- domain: artificial-intelligence, computer-vision, large-language-models
- platform: windows, python, cross-platform
- tags: computer-use, gui-agents, computer-automation, agent-computer-interface, mllm, osworld, in-context-reinforcement-learning, screen-understanding, ai-agents, automation, linux, macos

## Member repositories
- simular-ai/Agent-S (main) score 70

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:51.995862+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:14:50.930071+00:00, confidence not recorded.
  - readme: https://github.com/simular-ai/Agent-S (fetched 2026-08-28T04:10:51.995862+00:00, sha 2c55180db693)
  - homepage: https://www.simular.ai (fetched 2026-08-29T08:11:57.049257+00:00, sha a51dd723cf18)
  - site_page: https://www.simular.ai/about (fetched 2026-08-29T08:11:57.052319+00:00, sha 9f77b2ef49bd)
  - site_page: https://www.simular.ai/pricing (fetched 2026-08-29T08:11:57.054779+00:00, sha 17c5fe2a5666)
  - site_page: https://www.simular.ai/articles/2026-the-year-desktop-agents-stop-being-a-toy (fetched 2026-08-29T08:11:57.057190+00:00, sha e1bf06ecda74)
- Data as of 2026-08-30T08:39:29.467469+00:00.