Media
5 items
Description
自然语言驱动的浏览器 Agent：点击、输入、滚动、导航、提取，全部本地运行。 mio is a browser agent you talk to. Open the side panel, type what you want — "search for hello and click submit", "fill this form", "scroll and extract the article" — and mio plans the steps and executes them on the real page. When you open a page, mio suggests what can be done on it ("本页可做"), so you don't even have to describe a task to get started. It runs entirely in your browser: - Zero third-party dependencies (no SDKs, no bundles, all hand-written JS) - Bring your own LLM (OpenAI-compatible endpoint; point it at a local Ollama to stay fully offline) - No account required, no phone-home, no analytics. Page data goes only to the endpoint you configure. - Optional cloud sync (opt-in, account required) to back up task history across devices - Open source (MIT) on GitHub Built for: - Anyone automating repetitive web tasks (filling forms, checking pages, extracting data) - Developers and QA scripting multi-step flows - Privacy-conscious users who want a local agent What it can do: - Click, type, paste, scroll, navigate, switch tabs, extract text - Suggest tasks from the current page with one-click cards - Plan multi-step tasks with a recovery engine (finds alternate elements and retries instead of failing) - Remember context across steps and pages - Read CAPTCHAs by asking the human (never solves them in the background)