AI Agents
ESP32 iPhone Agent
ESP32 iPhone Agent — let an AI model use a real iPhone with a $7 ESP32: AirPlay eyes, USB HID hands, no computer, no jailbreak
★ 2.4k starsMIT
View guide →Jev Ultrafast is an open-source ai agents project by browser-use (browser-use/jev-ultrafast on GitHub). It has 23,674 GitHub stars and 1,794 forks, is released under the MIT license, and was last updated on GitHub on 30 September 2026.
Jev Ultrafast — the fastest browser agent from browser-use: one decision-model request per action cycle drives a real browser, MIT
Clone the repo, then let uv install dependencies — this also installs the Browser Harness that connects Chrome.
Copy .env.example to .env and add a TypeSafe API key and a text-model API key (OpenRouter in the example configuration).
Run the demo UI and click Start demo, or use the Agent library with your own URL and goal.
Operation and target heads share the same observed state, so each action costs one decision-model request, not a chain of them.
Jev consumes structured page state — visible controls, names, values and text — instead of screenshots; screenshots are an inspector opt-in.
Clicks check the document, form values, target and nearby context before acting, and covered controls are rejected before input.
The whole loop is small enough to read in full — the README points at jev_ultrafast/agent.py as the place to understand it.
The README route: clone, sync, configure keys, run the demo UI on http://127.0.0.1:8766.
git clone https://github.com/browser-use/jev-ultrafast.git
cd jev-ultrafast
uv sync
cp .env.example .env
uv run jevIf Chrome does not connect, the README route is the harness doctor command.
uv run browser-harness --doctorDrive any page from Python with a URL and a natural-language goal.
from jev_ultrafast import Agent
with Agent("https://www.google.com/travel/flights?hl=en", "Find one-way flights from Zurich to London, one adult, economy.") as agent:
for state in agent.run():
print(state["elapsed_ms"], state["status"])The browser-use organisation, built on TypeSafe's Jev decision model.
Yes — MIT licensed. You still pay for the API keys it calls (TypeSafe and a text model) under those services' own terms.
browser-use is the general browser-agent framework; Jev Ultrafast is the specialised fast loop on top of it — one decision-model request per cycle against an indexed action space.
No — the README is explicit that the flight example checks the route, date and results and saves its trace, but does not select or book a flight.
Clone, uv sync, add TYPESAFE_API_KEY and TEXT_MODEL_API_KEY to .env, then uv run jev for the demo UI.
ESP32 iPhone Agent — let an AI model use a real iPhone with a $7 ESP32: AirPlay eyes, USB HID hands, no computer, no jailbreak
Snowflake's Agent World Model: synthesise 1,000 executable MCP tool-use environments for agentic RL.
The open-source library that lets AI agents control a real browser.
ZCode — Z.ai's AI coding workspace: desktop app, browser interface and terminal agent in one open-source harness
Reliable computer use for any agent: drive real apps through accessibility trees, not pixel guessing.
OpenMausBot — the open-source Grok Bot alternative: your own team of AI bots in a chat app, each a real local agent with its own personality, model and cloud computer
The most-starred other AI Agents projects in the LibreAI catalogue — the realistic shortlist if Jev Ultrafast is not the right fit.
A personal AI that really does things — on any OS, with your data staying yours.
An agentic skills framework & software development methodology that works.
Agent harness optimization system with skills, memory and security for Claude Code, Codex and more
The agent that grows with you — a Python framework for adaptive AI agents.
DeepSeek's plugin-based agent harness where everything is a plugin.
Nearby AI Agents projects in the catalogue, closest in stars to Jev Ultrafast — keep exploring the shelf either side of it.
The official Python SDK for Model Context Protocol servers and clients.
One RAG framework for everything — including multimodal documents.
Stateful AI agents with memory that learns and improves.
A framework for building agents and services
Autonomous AI agents for penetration testing — self-hosted, multi-agent, with a full web UI
An MCP server that lets Claude, Cursor and Windsurf build n8n workflows for you
Free and open — grab the source, star the repo, and start creating.