GitHub radar
Describe a task in plain English — the agent clicks for you
This is a browser agent: you give it a goal in plain language, and it finds the right buttons and fields on a webpage and acts on them — no site-specific code required.
A program opens a website in a browser, reads the visible elements — buttons, input fields, dropdowns — and clicks or types step by step to reach your goal.
When it helps
Useful for automating actions on a site you don't control the code of: search, booking, forms. Not for complex general-purpose tasks, and don't trust it to finish a booking or payment unchecked.
Pros
- One network request per step — runs fast
- No GPU needed, runs on a regular laptop
- No per-site code — you just describe the goal as text
Cons
- Needs paid API keys for TypeSafe and OpenRouter
- Not built for complex general-purpose tasks
- Don't trust its "done" report without checking yourself
How to set it up — step by step
- 1Open your AI agent (Claude Code or Codex) in your project folder.
- 2Send the agent the text from the block below.
- 3Ask the agent to add TYPESAFE_API_KEY and TEXT_MODEL_API_KEY to the .env file — give it the keys if it asks.
- 4Start the project and open http://127.0.0.1:8766 in your browser.
- 5Click "Start demo → Run automatically" and watch it search Zürich–London flights.
- 6Check the result by hand — it doesn't finish bookings and can make mistakes.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up this browser agent: https://github.com/browser-use/jev-ultrafast — follow the README, add TYPESAFE_API_KEY and TEXT_MODEL_API_KEY to .env (ask me for keys if needed), run it and open http://127.0.0.1:8766, then run the flights demo.
Runs on any laptop — only Python needed; no GPU required, but paid TypeSafe and OpenRouter API keys are necessary.
Open on GitHub▌ More finds