User repository on GitHub
Jev-driven browser agent in pure Go — CDP over pipe/WebSocket, attach to real Chrome, built to measure real speed
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Tools & CLIs
Browser AI
A browser agent in pure Go (stdlib only) where Jev picks each step from a finite indexed action space over Chrome CDP — headless, attached to your real Chrome, or relayed through an unpacked extension. Every step is instrumented for timing.
- Stars
- 1
- Language
- Go
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- phanngoc
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Languages
- Go
- Platforms
- BrowserCLIChrome extension
- Capabilities
- Open sourceRequires API key
Highlights
- Pure Go, stdlib only: CDP over --remote-debugging-pipe or a minimal WebSocket client attached to your running Chrome.
- Jev picks one operation and one element index per step; a small OpenAI-compatible model writes text only for TYPE_TEXT.
- Benchmarks on 2026-09-20, headless Chrome 153: Jev 3/3 passes on Google Flights in 9.6 s median and 3/3 on Wikipedia in 3.4 s.
- Of a 9.6 s Flights run: Jev 6.0 s across 18–19 decisions, text model 1.2 s, browser side 1.2 s over ~220 CDP calls.
- About 330 ms per decision from Vietnam, ~200 ms of it network; Jev itself is about 130 ms.
Quickstart
bash
cp .env.example .env
go run ./cmd/agent --headless --url https://en.wikipedia.org/wiki/Main_Page --goal "Find and open the Wikipedia article about Gödel's incompleteness theorems."Watch out
Copy reviewed against the truncated repo README on 2026-10-15; the excerpt stops inside the extension section. Pass tables are the project's own runs on one machine on 2026-09-20, and the README itself notes the reference Python implementation reports different European timings.
Questions about this entry
Short answers drawn from the linked project or page.
A browser agent written in pure Go (stdlib only) and driven by Jev. It attaches to Chrome over CDP — launched headless with --remote-debugging-pipe or attached to your real running Chrome over a minimal WebSocket client — and instruments every step so you can see where time goes.
Jev is given a finite, indexed action space and picks one option per step, returning a probability over each; it does not generate text. A small OpenAI-compatible LLM writes field values only when the chosen operation is TYPE_TEXT.
On the README's 2026-09-20 runs, Jev passed Google Flights 3/3 in 9.6 s median and Wikipedia 3/3 in 3.4 s; gemini-2.5-flash also passed Flights 3/3 but took 23.0 s, gpt-5.4-mini passed 2/3 with reasoning on, and the small models never passed.
cp .env.example .env and fill TYPESAFE_API_KEY plus TEXT_MODEL_API_KEY (OpenRouter by default) for the text model that writes TYPE_TEXT values. The README's quick start then runs go run ./cmd/agent --headless with a --url and --goal.
Yes. --attach auto attaches to your running Chrome so it has your profile, cookies and logins, and --via-extension relays CDP through a small unpacked extension loaded once in chrome://extensions, so no debugging port is needed.
The Python reference — browser-use/jev-ultrafast (MIT) — is where the DOM snapshot script and policy instructions come from. This repo is a Go reimplementation; the README projects ≈6.5 s for the same Flights task from the same continent as the Python reference's 7.09 s.
Small models (gemini-2.5-flash-lite, gpt-5.4-nano) failed the Flights task every time, sometimes declaring DONE early, and gpt-5.4-mini needed reasoning on to pass. All figures are the project's own benchmark runs, and the pasted README stops inside the extension section.
More Jev resources like this
A Go linter that checks code against plain-language rules: Tree-sitter extracts code units, then each one is sent to Jev or another configured decision provider to decide whether it passes, with caching and eval fixtures.
Jev Tools & CLIs#community#catalog-import#logicrw
Community
codegirl-007/jevlintGo
codegirl-007/jevlintGoWrite end-to-end tests for web apps in plain English; jev-e2e turns each case into a plan, has Jev pick the page controls, and runs it with Playwright, returning PASS, FAIL or BLOCKED with HTML, JSON and masked screenshot evidence.
Jev Tools & CLIs#community#catalog-import#madewithjev
Community
perixtar/jev-e2eTypeScript
perixtar/jev-e2eTypeScriptA browser automation CLI for AI agents: drive Chrome with natural language in an interactive terminal or structured commands like jevb page act, signed in with a TypeSafe or OpenRouter key. Prebuilt packages target macOS Apple Silicon; Node.js 22+.
Jev Tools & CLIs#community#catalog-import#logicrw
Community
mrlyk/jev-browserRust
mrlyk/jev-browserRustA pre-action gate for tool-using agents that decides allow, confirm or block with deterministic rules before a neural decider, plus a CDP browser loop whose decider swaps between hosted Jev and a local System One-compatible server through environment variables alone.
Jev Tools & CLIs#community#catalog-import#yibie
Community
pusucip25/local-deciderPython
pusucip25/local-deciderPythonRemote control for coding agents: give Claude Code, Codex, Cursor, OpenCode, Pi and others a task from your phone, and get a notification when the result is ready while your laptop stays closed. Self-hosting is a free MIT install; the hosted version has a free plan and paid plans from $19/mo.
Jev Tools & CLIs#community#catalog-import#logicrw
Community
bennykok/omg.devTypeScript
bennykok/omg.devTypeScriptA customizable linter that asks decision models like Jev and Clef your fuzzy rules — does the comments in this file repeat what the code says? — in a single Go binary with no runtime and minimal dependencies.
Jev Tools & CLIs#community#catalog-import#madewithjev
Community
madxmike/gallantGo
madxmike/gallantGoFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.