Organization repository on GitHub
Fastest and cheapest web agent
- Forks
- 1.7k
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
Jev directory
CommunityJev Tools & CLIs
Jev Ultrafast
Browser Use's ultrafast agent: Jev picks an operation and an element from an indexed action space in one request, and a small LLM writes text only when the action is TYPE_TEXT.
- Stars
- 23.2k
- Language
- Python
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- browser-use
- Last verified
- Sep 2026
About this project
A browser agent with a dynamic, indexed action space. Give it one goal. TypeSafe's Jev picks an operation and an element. A small LLM writes text only when the operation is TYPETEXT. Zürich → London on Google Flights in 7.1 seconds. One natural-language goal, actual text generation, and loading waits included. Watch the MP4 · Measurements · Read the loop. Every observation produces a new element table:
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
Highlights
- One TypeSafe request returns the operation plus every target head, and only the target matching that operation executes.
- The measured demo runs a Zürich to London Google Flights search in 7.1 seconds, including text generation and loading waits.
- No screenshots in the default loop: Jev consumes an element table read atomically from the DOM.
- Wait budgets are explicit: up to 200 ms for combobox suggestions, otherwise two animation frames or 50 ms.
Quickstart
python
from jev_ultrafast import Agent
with Agent(
"https://www.google.com/travel/flights?hl=en",
"Find one-way flights from Zurich to London on September 20, 2026.",
) as agent:
for state in agent.run():
print(state["elapsed_ms"], state["status"])Watch out
MIT-licensed. Needs a TypeSafe key plus a separate text-model key (OpenRouter in the example); tabs, nested scrolling, and arbitrary keyboard widgets are outside this MVP.
Questions about this entry
Short answers drawn from the linked project or page.
Both numbers come from six alternating runs of a single task on one browser profile: three of three passes for each version, median task time 9.450 s to 7.092 s, and median browser protocol calls down from 1,092 to 101. The README calls that three repeats of one task, not a general reliability benchmark.
Not yet. The DOM reader handles common HTML and ARIA controls rather than the full accessible-name specification, and shadow roots, frames, canvas, uploads, pop-up tabs, nested scrolling, and arbitrary keyboard widgets are named as outside this MVP. Expect to pair it with a conventional agent on those sites.
You need both keys. Jev picks the operation and the element, but a small LLM writes the text, and only when the chosen operation is TYPE_TEXT. The example configuration uses an OpenRouter key for inception/mercury-2.5 with reasoning disabled; Gemini, GLM, and DeepSeek can also use the OpenAI-compatible text helper.
Model output never becomes selectors, coordinates, shell commands, or executable JavaScript. Every executed target resolves from an observed node, the executor rechecks page freshness and click occlusion, and text-helper output must parse as a small JSON object before typing. One thing to know: owned tabs share your existing Chrome profile.
Clone the repo, `uv sync`, copy .env.example to .env and add TYPESAFE_API_KEY and TEXT_MODEL_API_KEY, then `uv run jev`. Open http://127.0.0.1:8766 and click Start demo, then Run automatically. Chrome connects through Browser Harness, installed by uv sync.
CLICK, TYPE_TEXT, SELECT, SCROLL_UP, SCROLL_DOWN, WAIT, DONE and BLOCKED, with only supported operations and targets offered. Target questions are speculative: if the operation is CLICK, only click_target can execute, and both decisions share one network round trip.
Import Agent from jev_ultrafast, construct it with a start URL and goal, and iterate agent.run() states, each carrying elapsed_ms and status. Run it with `uv run --env-file .env python your_script.py`. A DONE choice still needs independent outcome verification.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: Jev Ultrafast: a browser agent with an indexed action space
rahimnathwani92 points · 14 commentsJev Ultrafast: A browser agent with a dynamic, indexed action spaceRead the thread on Hacker NewsX: Text-to-screenshot (Malay Vasa)
X: Google's MediaPipe Decision Maker, tested in the browser
X: Playwright CLI + Jev vs Playwright MCP
More Jev resources like this
Multi-agent trading research with optional Jev screening of StockTwits and Reddit posts for company relevance before sentiment analysis runs.
Jev Tools & CLIs#community#python#trading
Community
TauricResearch/TradingAgentsPython
TauricResearch/TradingAgentsPythonAn open-source browser agent extension for Chrome and Edge that browses, fills forms, clicks, and collects data, letting any extension-capable browser act as an AI browser with your own LLM key.
Jev Tools & CLIs#community#typescript#browser
Community
dom-actions/domaTypeScript
dom-actions/domaTypeScriptA hands-free macOS voice assistant: whisper.cpp transcribes locally in about 100 ms, then one roughly 250 ms Jev request selects a typed action plus arguments, and code executes it against the Mac.
Jev Tools & CLIs#community#python#voice
Community
kevinbadi/jev-voicePython
kevinbadi/jev-voicePythonA TypeScript browser agent for ego lite that turns one natural-language goal into CLICK, TYPE_TEXT, or SELECT on a numbered element table, with Jev choosing the operation and target in one request.
Jev Tools & CLIs#community#typescript#browser
Community
romaluev/jev-egoTypeScript
romaluev/jev-egoTypeScriptA cross-platform memory layer for AI coding agents: conversations from Claude Code, Codex, DeepSeek Harness, OpenClaw, and OpenCode become Markdown memories indexed in Milvus, with optional Jev reranking through the TypeSafe API.
Jev Tools & CLIs#community#python#agents
Community
zilliztech/memsearchPython
zilliztech/memsearchPythonA desktop AI assistant with a Jev / System One inline plugin for classification, routing, verification, and scoring without chat-style output.
Jev Tools & CLIs#community#python#desktop
Community
szczyglis-dev/py-gptPython
szczyglis-dev/py-gptPythonFrom 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.