Most browser agents send a screenshot to a large model at every step. The projects here read the page or accessibility tree deterministically, turn the possible actions into a list, and ask Jev to pick one. A generative model is only called when text actually needs to be typed.
The result is far fewer expensive model calls per task. The trade-off is that Jev can only choose among options your code puts in front of it.
browser-use/jev-ultrafast
Fast browser agent from Browser Use. Jev decides each step and which element to act on; a small model is called only when text needs to be typed. The authors report a full Google Flights search in about 7.1 seconds.
vercel-labs/agent-desktop
macOS desktop automation over accessibility trees. Since v0.9.2, its jev-desktop scripts let Jev choose which control to operate and which action to take.
awlevin/typesafe-computer-use
macOS computer use without sending screenshots to a large model. The screen is read deterministically and Jev picks the next action.
wy-coliney/jev-browser-use
Browser skill that splits the work: Jev handles navigation, clicks, toggles and scrolling while the coding agent thinks and verifies. Uses your existing browser connection, with no extra driver.
Sac-Y/Jev-cu
Codex computer-use skill where Jev picks the next element and action from on-screen text, with a local policy gate for sensitive steps. README in Chinese.
droidrun/mobile-jev
Android phone automation from DroidRun on its Mobilerun device cloud, with Jev choosing every operation and target. The author reports reaching Uber's payment selection in about 21 seconds.
moritzkremb/jev-voice-browser
Voice-controlled Chromium: on each partial transcript, one Jev request judges intent, target element, and whether the command is complete or destructive.
jkudish/jev-browser
Headless browser automation through an MCP server, CLI, or library. Jev picks one action per step.
shhivv/arc-cua
Lets a planner hand bounded desktop subtasks to a decision model that runs the UI loop, with a lean macOS driver usable on its own.
aindeev/agent-chrome-relay
Lets coding agents use your signed-in Chrome on macOS in background tabs, carrying out a whole goal per command with Jev (or Claude Haiku through Claude Code) choosing only the steps it cannot fill or replay. With Jev on, page text of agent tabs goes to TypeSafe.
socai-io/jev-social
Instagram, TikTok and LinkedIn research where Jev chooses each next operation and a real Chrome session executes it, with the evidence kept.
KYRIE66nb/mobile-agent
On-device Android agent that reads and taps any app through accessibility and vision; its allow / confirm / block safety gate can be handed to Jev or a self-hosted Laya over /v1/systemone, off by default. In Chinese.
agent-labs-dev/fastbrowse
Browser agent that indexes the page into candidates for Jev to pick from, leaves planning and reading to an LLM, and cites a quote from the page for every claim.
YUTA-fywoo/jev-gui-delegate
Windows and Chrome GUI delegation for Codex: the model hands over a task contract, a local controller runs deterministic steps, and low confidence stops for a person. Chinese.
timpratim/macbrow
Say a command and it runs as AppleScript, or say a web task and it drives Chrome. Its README opens with a warning about an early version tidying a Desktop rather thoroughly.
kevinbadi/jev-voice
Local whisper.cpp for the transcript, then one Jev request picks the action and its typed arguments; code owns execution.
savka777/jev-use
macOS computer use by voice or typing that reads the screen through the Accessibility tree rather than screenshots. Key stored in the Keychain.
michaelswissa/jevry
Desktop browser agent where a language model plans, a decision model picks the action, and Chromium carries it out.
yikangy873-gif/jev-desktop
Adds a bounded decision loop to Codex Computer Use for browser tabs and native macOS apps.
brainstormity/System-One-Computer-Use
Prototype screenshot loop that asks Jev for the next click, keystroke, scroll or site to open and carries it out with pyautogui or on an Android phone over adb, with a local dashboard that records each run. It controls the real mouse and keyboard.
buer2233/jev-ui-test
UI test framework where cases are written as sentences and each step is scored against the page's candidate elements rather than generated, run by pytest with Allure reports. Chinese.
dushyantzz/Ghosthand
Windows desktop assistant: focus an app, press Ctrl+Win and describe the task; it reads UI Automation controls and has Jev choose each action through Vercel's AI Gateway, with human-in-the-loop checks.
chy4pro/jev-for-chrome
Manifest V3 port of jev-ultrafast that drives the tab you are already looking at, keeping the same observation format and execution rules.
hqman/jev-browser-skill
Agent skill that drives an isolated Playwright Chromium: the agent sets a narrow goal and Jev chooses the in-page actions. Vercel AI Gateway by default, TypeSafe optional.
ronadin2002/jev-cua
Floating bar for macOS that takes a voice or text command and picks the next action from the Mac's live accessibility controls, looping until the goal is met.
hqman/JevScout
Coding-agent skill that drives Chrome over CDP to look for jobs on company sites, with Jev scoring pages and links.
GoldenLoaf24h/browserpaw
Drives your everyday Chrome from an agent across fifty MCP tools, with each step decided by Jev or by a decider model it downloads and runs on your own machine.
mhrsdev/zero-use-computer
Computer use over MCP that reads what a screen reader reads and acts on real controls, with an optional decision model - Jev or any compatible server - for the small choices.
jiangkoumo/ego-decision-layer
Replaces the per-step LLM turn in the ego lite browser with one Jev call that picks both the action and the target element, while code observes, executes and fails closed. The installer takes over the official ego-browser skill entry unless you pass --no-wire.
bgivenb/flick-computer-use
Local stdio MCP server that executes a whole browser or macOS goal from the agent's goal, values and completion condition.
jcpsimmons/jev-macos-loop
Native macOS GUI automation on Apple silicon: OmniParser CoreML and Apple Vision OCR identify controls locally, Jev selects the action. AGPL-3.0.
ChenneyZhuang/laya-browser-agent
Browser agent whose next action is decided by Laya on your own machine, with no cloud and no API key. Japanese.
jiawei686/jev-ultrafast-mcp
MCP server that takes a whole browser task in one call and drives the page itself, so the agent never opens a browser.
filedcom/playjev
Natural-language browser automation for Playwright, in the shape of Stagehand. Not the same project as the PlayJev decision model below.
IAmJSD/OpenComputerUse
Background computer use for agents as an MCP server, written in Rust: sessions drive an app behind your other windows, and plain-English recipe steps are placed by one Jev (or Clef) choice over the window's interactive elements, stopping when it is not confident.
forvela/jev-agent-browser
Bounded browser tasks for a parent agent: Jev picks the next typed action, agent-browser runs it, and ambiguity or a stuck state comes back as a structured handoff.
Ice-Hazymoon/jevwright
Browser tests for business flows written as the steps a user takes: the model finds each control once, and the recording replays without it.
ironbee-ai/ironbee-express
Browser agent for checking that a web app really did what the page says: each step is one Jev choice over the page's controls, a text model is optional, and a recorded run replays with no step decisions. Elastic-2.0.
tontoko/jev-browser
One core behind a typed SDK, a persistent CLI and an MCP server: native Playwright operations and assertions need no key, and only the natural-language paths call Jev. Japanese.
KiritoKing/midscene-jev-runner
Community runner for Midscene Test that adds jevAct, jevAssert and jevWaitFor nodes over a Playwright page, with a bounded non-text action space: click, select, scroll, wait, dismiss.
Ryu0118/jev-sim-use
Drives an iOS Simulator or Android device without spending a frontier reasoning turn on every tap.
idovmamane/dejevu
Browser agents that act on one look at the page; the default backend is any open model, and --backend typesafe makes Jev the chooser instead.
jacks3tr/Jev-Desktop
Lets an agent hand off a Windows desktop task through MCP or the command line: Jev observes windows, chooses controls and types supplied text, and the agent gets the result or a request for help. Alpha.
xinwang-nwpu/jev-mobile
Operates an Android phone from one sentence: Jev picks one action per step from the accessibility tree, executed over ADB, with a local or vision model available to take over. Chinese and English.
danielnc/jev-browse
Helpers for browser-harness that let Claude Code or Codex hand off a whole website sub-task in one call, with Jev deciding each step and returning the outcome or a typed reason it stopped. A port of jev-ultrafast.
openqa-cn/jev-browser
Browser automation that indexes the controls inside the page and has Jev choose one, instead of sending a screenshot to a vision model on every step.
New Jev releases, pricing changes, and the best new projects, once a week. No spam; unsubscribe anytime.
Powered by Buttondown. See our privacy policy.