Auto Web Agent

Control your browser with natural language: auto-click, type, navigate, extract. Works with your own custom AI models.

Auto Web Agent lets you control your browser in plain language — describe a goal, and it does the clicking, typing, navigating, and content extraction for you. ■ What it's for Turn any Chrome tab into an AI assistant. Just give a task in natural language — e.g. "Search Bing for Chrome extension development and open the first result", "Summarize the main points of this page", or "Open github.com and search for nanobrowser" — and the agent watches the page, decides the next step, and acts on it, over and over, until the job is done. ■ Why you should install it · Natural language, not scripts: describe the goal; the agent figures out the steps. No more repetitive clicking. · Fully transparent: a live side panel shows each step's thought → action → result, and the element it's about to touch lights up on the page with a colored, numbered box. · Bring your own model, no cloud lock-in: works with OpenAI, DeepSeek, Claude, OpenRouter, local Ollama, or any custom endpoint (API Key / Model / Base URL / Headers). · Privacy-first: your API key and settings stay in local browser storage; page content is sent only to the model you configure, only while a task runs. No analytics, no trackers, no third-party servers. · Bilingual: the UI and the agent's reasoning switch between English and 中文. · You stay in control: stop anytime, cap the number of steps. · Featherweight: pure JS/CSS/HTML, zero build step. Great for researchers, QA testers, developers, and anyone tired of repetitive clicking. How to use: after installing, open the side panel → enter your model's API key and model name in Settings → open any page and type a task. Auto Web Agent 让你用一句话指挥浏览器——描述目标，它自己完成点击、输入、导航和内容抓取。 ■ 用途 把任意 Chrome 标签页变成一个 AI 助手。你只需用自然语言下达任务，例如「在必应搜索 Chrome 扩展开发并打开第一条结果」「总结这个页面的主要内容」「打开 github.com 并搜索 nanobrowser」，Agent 会自动观察页面、决定下一步动作并逐步执行，直到任务完成。 ■ 为什么值得安装 · 自然语言，无需脚本：描述目标即可，Agent 自己拆解步骤，告别重复点击。 · 过程完全可见：侧边栏实时展示每一步的「思考 → 动作 → 结果」，页面上用彩色编号框高亮正在操作的元素。 · 自带模型，不锁定云端：支持 OpenAI、DeepSeek、Claude、OpenRouter、本地 Ollama 或任意自定义端点，可自定义 API Key / Model / Base URL / Headers。 · 隐私优先：API Key 与设置仅保存在浏览器本地，网页内容只在任务运行时发送给你自己配置的模型，无分析、无追踪、无第三方服务器。 · 中英双语：界面与 Agent 的回答语言一起切换。 · 完全可控：随时一键停止，可限制最大步数。 · 轻量零构建：纯原生 JS/CSS/HTML，加载即用。 适合研究人员、QA 测试、开发者以及所有厌倦重复点击的人。 使用方法：安装后打开侧边栏 → 在设置中填入你的模型 API Key 与模型名 → 打开任意网页，输入任务即可。