AIエージェント
steel-browser
A batteries-included browser API so agents can automate the web.
★ 7.8k スターApache-2.0
ガイドを見る →webcmdは、agentrhqが公開しているオープンソースのAIエージェントプロジェクトです（GitHub: agentrhq/webcmd）。GitHubスターは2,665、フォークは1,542で、Apache-2.0ライセンスの下で公開されています。GitHubでの最終更新は2026年9月25日です。
Self-learning browser infrastructure for AI agents — sites remembered, so agents stop rediscovering the web
Name the outcome you want from the web — research with source links, data from a logged-in site, a price/availability check — in one prompt.
Install the npm package globally (Node.js 20.6+ required).
Run webcmd skills add and choose your harness (Claude, Codex, another supported harness, or a custom path).
Tag the webcmd-browser skill for live browser work and describe the outcome; Webcmd works in a real browser and quietly retains navigation context for next time.
Agents inspect, click, type, extract and capture network calls in a real browser to complete tasks on unfamiliar sites.
For familiar sites, Webcmd keeps an agent-facing sitemap of observed pages, states, actions, workflows, APIs, pitfalls and fallback paths.
Profiles are cookie jars for authenticated sessions (e.g. a 'work' or 'social' profile); sessions are independent browser windows within a profile, so parallel agents can work safely.
For multi-step local exploration, agents can send one Playwright-style program to an explicit browser session via webcmd browser run.
After an optional first-access cloud seed, subsequent learning stays local; learning never blocks the task and the live browser is always the truth.
Used across research/community sites, social and professional networks, AI tools, shopping and booking products — the README stresses the same workflow works on other sites too.
Clone webcmd if you want the source; the normal install is the npm package in step 2.
git clone https://github.com/agentrhq/webcmd.git
cd webcmdInstall globally and add the browser skill to your harness, as the quick start shows.
npm install -g @agentrhq/webcmd
webcmd skills addCreate a named session in a profile, inspect its tabs, and close it when done (raw browser commands need an explicit session ID).
webcmd --profile work session create "Work Project" -f json
webcmd --profile work session close work-project-k7agentrhq (AgentR) maintains it; docs are at webcmd.dev.
The package is Apache-2.0 licensed open source. Webcmd Cloud (hosted sessions) is described as in active development and not yet stable.
Webcmd is browser infrastructure your agent drives; model costs belong to the harness (Claude, Codex, etc.) you connect it to.
Those start from zero on every run. Webcmd adds a persistent, self-learning memory of how sites work, so repeat tasks skip rediscovery.
npm install -g @agentrhq/webcmd, run webcmd skills add, then ask your agent for a web outcome using the webcmd-browser skill.
A batteries-included browser API so agents can automate the web.
PraisonAI 🦞 — Hire a 24/7 AI Workforce. Stop writing boilerplate and start shipping autonomous self-improving agents that research, plan, code, and execute t…
Reef — infrastructure for continually self-improving agents: weight-training and harness-evolving deployments.
Autonomous AI agents for penetration testing — self-hosted, multi-agent, with a full web UI
Bright Data's MCP server: web search, scraping, structured data and browser automation for agents (75 tools)
Hook and skill guard that stops coding agents from doing unrequested busywork.
LibreAIカタログに掲載されたAIエージェントプロジェクトの中で最もスターの多いもの — webcmdが合わなかった場合に検討したい、現実的な候補リストです。
A personal AI that really does things — on any OS, with your data staying yours.
An agentic skills framework & software development methodology that works.
Agent harness optimization system with skills, memory and security for Claude Code, Codex and more
The agent that grows with you — a Python framework for adaptive AI agents.
DeepSeek's plugin-based agent harness where everything is a plugin.
カタログの中でwebcmdとスター数が近いAIエージェントプロジェクト — 前後の棚もそのまま見て回れます。
Python library for building multimodal language agents — VLM, video and mobile, fast to prototype and ship
Bright Data's MCP server: web search, scraping, structured data and browser automation for agents (75 tools)
Free Chinese AIGC course: prompting, agents, image, video, voice, open models.
NVIDIA's toolkit for connecting, profiling and optimizing teams of AI agents — across any framework
YAO: rigorous engineering, evaluation, and governance for reusable agent skills.
Rust terminal multiplexer with a typed SDK: drive any CLI app from code.