More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Jev Browser

    1

    Let Jev drive a real Chrome: name a site and a goal, and it clicks, types and selects its way there at a few hundred ms per step.

    Free

    1 useSecurity scanned
    Jev Browser

    Jev Browser

    Example session with this skill installed

    Use Jev to search English Wikipedia for Alan Turing and open his article. Leave the page open and show me the steps and timing.

    • Read your context and instructions
    • Compiled the jev browser

    Opened Alan Turing's article on English Wikipedia and left the browser page open.
    The task finished successfully in 10.3 seconds.

    Final page

    The visible introduction describes Alan Turing as an English mathematician, computer scientist, logician, cryptanalyst, philosopher, and theoretical biologist. It discusses his work on theoretical computer science and the Turing machine.

    Actions and timing

    StepActionTarget or resultJev decision timeElapsed
    1TypedSearch Wikipedia: Alan Turing495 ms8.6 s
    2ClickedAlan Turing article suggestion242 ms9.2 s
    3FinishedArticle visible197 ms10.3 s

    Total Jev decision time: 0.934 seconds. The total elapsed time also includes browser startup and page interaction.

    Next steps

    • Read the article in the browser window left open.
    • Ask me to save a screenshot of the current page.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Let Jev drive your browser. You name a site and a goal; Jev, TypeSafe's single-forward-pass decision model, clicks, types and selects its way there in a real Chrome window while you watch. A few hundred milliseconds per decision.

    Source code: github.com/zurfyx/jev-browser-skill
    Step through a real run (interactive explainer): jev-browser.vercel.app

    Jev logging in to Hacker News, searching for Jev and opening the top thread, at 1x speed

    What it does

    • Searches websites: enters your search terms and opens the result that matches your goal.
    • Fills forms: types the values you provide, picks dropdown options, ticks checkboxes and submits.
    • Logs in safely: a --secret is typed only into password fields and is never sent to Jev or printed.
    • Shows its work: every step prints the operation, the target, Jev's probability and its latency, then a JSON summary with the final URL, title and visible page text.
    • Uses your browser or its own: drives your Chrome (tabs and logins included) when remote debugging is allowed, otherwise a reusable Jev window whose logins persist between runs.

    How it works

    1. Observe. A snapshot runs inside the page and returns a numbered table of the visible controls. Jev never sees the DOM.
    2. Ask. One request asks Jev for the next operation (CLICK, TYPE, SELECT, ENTER, DONE, BLOCKED) and a target for every operation at once, so two dependent decisions cost one round trip.
    3. Validate. The code reads only the target matching the chosen operation. Any answer outside the offered set throws before anything runs. Model output never becomes a selector or code.
    4. Execute with real mouse and keyboard events, then observe again until Jev answers DONE or BLOCKED.

    Measured speed

    TaskStepsTotalTime in Jev
    Wikipedia: search and open an article23.4s1.0s
    Selenium web form: two fields, a dropdown, a checkbox, submit55.1s1.4s
    Hacker News: log in, search, open the top thread's comments79.3s2.4s

    A step is 3 to 6k input tokens and output is free, so the whole Hacker News run costs less than a tenth of a cent.

    Built to be read

    A reference implementation: three short files, six operations, no dependencies (no npm install, no Playwright, no browser download). scripts/core.mjs holds the whole decision logic in about 250 lines, and the explainer site replays recorded traces step by step, or runs the loop live with your key. Full docs are in the GitHub README.

    Requirements

    • Node 22+
    • Chrome, Edge, Brave, Arc or Chromium
    • A TypeSafe API key in TYPESAFE_API_KEY (environment or a .env file in the skill folder)

    Known limitations

    • Jev chooses, it never generates: every string to type must be supplied up front.
    • At most 250 controls per page are offered (Jev takes up to 255 choices per question). For dense pages, start from a URL closer to the target.

    MIT licensed. The action-space design follows browser-use/jev-ultrafast.

    Requires Node.js 22+, a Chromium-based browser, and a TypeSafe API key. Typing needs supplied text. Only the first 250 page controls are considered; tasks can stop unfinished.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Permissions

    Terminal / Shell
    Browser
    Network Access
    Environment Variables

    Shell: runs `node scripts/jev.mjs`. Browser: drives Chrome over the DevTools Protocol (your own Chrome only if you allow remote debugging, otherwise a separate Jev window). Network: one request per step to api.typesafe.ai with the page's control table and visible text; passwords passed via --secret are never sent. Env: reads TYPESAFE_API_KEY (or a .env file in the skill folder). Source: https://github.com/zurfyx/jev-browser-skill

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Jev Browser automates web navigation by driving a real Chrome window to click, type, and select elements in real-time. It is specifically designed for speed, executing decisions in a few hundred milliseconds to handle tasks like form filling, logins, and multi-step site searches.

    Jev Browser works with any Chromium-based browser (Chrome, Edge, Brave, Arc) on a machine with Node 22+ installed. It can either drive your existing browser instance—maintaining your active tabs and logins—or run in its own dedicated, reusable window.

    The skill includes a 'safe login' feature where sensitive data provided via the `--secret` flag is typed directly into password fields by the driver. This ensures your credentials are never sent to the AI model or printed in logs, maintaining high security during automated sessions.

    Setup is lightweight with no external dependencies like Playwright or heavy npm installs required. You simply need a Node 22+ environment, a supported Chromium browser, and a TypeSafe API key configured in your environment variables.

    Jev Browser is unique because it uses a single-forward-pass decision model that predicts both the operation and the target simultaneously. This architecture significantly reduces latency and API costs, allowing complex tasks like navigating Hacker News to be completed in seconds for a fraction of a cent.

    Currently, Jev Browser is limited to a maximum of 250 interactive controls per page. If you are navigating a very dense website, it is recommended to start from a URL closer to your final goal to ensure all necessary elements are captured in the snapshot.

    New in Workflow Automation