Read & learn
Written guides.
Understand Jev, one idea at a time. Walkthroughs, recipes, and writeups — our notes first, the original next.
Explore the Jev ecosystem.
Learn, build, and play with Jev — from deep technical guides to creative experiments.
Watch & learn
Video guides.
See an idea take shape. Tutorials, demos, and deep dives, organized by topic and credited to their creators.
Make & explore
Experiments.
See what builders made with Jev. Games, tools, repositories, articles, and more — traced to their sources.
The directory03 / 03 · Experiments
Find something worth exploring.Experiments
Every category, from games and tools to repositories and writeups, traced to a primary source.
Real-time research powered by
Same ideas.
Different paths.
This is the full registry, not just the featured picks. Figures like speed, cost, and stars are maker-reported or captured snapshots, not JevMade measurements.
118 experiments · showing 1–60
Browser Use's agent asks Jev to choose the next browser operation and target element from a fresh table on every step.
SOURCE SCREENSHOTEXPAND ↗
from the source
git clone https://github.com/browser-use/jev-ultrafast.gitcd jev-ultrafastuv synccp .env.example .env# Add TYPESAFE_API_KEY and TEXT_MODEL_API_KEY.
Demonstrates a bounded Jev action chooser inside Cua Driver loops, with mock and credentialed live paths.
SOURCE SCREENSHOTEXPAND ↗
Drive a Mac through its accessibility tree, with Jev choosing an operation and a live element on each turn.
ACTUAL RECORDING12 SEC ↗
from the source
const req = buildRequest("save the file", { app: "TextEdit", window: "Untitled" }, offer...assert.deepEqual(Object.keys(req.questions), ["target", "command", "present", "destructi...assert.equal(Object.keys(req.questions.target.criteria).at(-1), "none");assert.equal(req.questions.present.type, "noul");assert.equal(req.questions.target.criteria[`@${S}:e3`].what, 'button "Save"');
A macOS menu-bar companion that turns a typed click task into a short sequence over locally detected controls.
SOURCE SCREENSHOTEXPAND ↗
from the source
"absent": .noul( instructions: "Is the control needed to make the next bit of progress on... ), "kind": .choice( instructions: "How should that element be acted on?",
A cautious Codex computer-use skill that chooses UI targets from accessibility text while local policy guards consequential actions.
SOURCE SCREENSHOTEXPAND ↗
A Mac menu-bar assistant that uses Jev to choose what to click.
SOURCE SCREENSHOTEXPAND ↗
from the source
nonisolated static let maxChoices = 255 nonisolated static let doneThreshold = 0.70 nonisolated static let absentThreshold = 0.50 nonisolated private static let noneKey = "__none__" endpoint: URL = URL(string: "https://api.typesafe.ai/v1/systemone")!) {
A macOS automation loop that reads the screen with OCR, asks Jev what to do next, and clicks the chosen target.
SOURCE SCREENSHOTEXPAND ↗
from the source
git clone https://github.com/awlevin/typesafe-computer-usecd typesafe-computer-useuv synccp .env.example .env # fill in the keys
Port browser-use's Jev-driven browser agent to local Laya on Apple Silicon while retaining an optional hosted Jev mode.
SOURCE SCREENSHOTEXPAND ↗
A mobile automation demo where Jev navigates Uber from route entry to payment selection.
ACTUAL RECORDING12 SEC ↗
from the source
git clone https://github.com/droidrun/mobile-jev.gitcd mobile-jevcorepack enablepnpm install --frozen-lockfilecp .env.example .env.local
Indexes visible browser controls, lets Jev choose an allowed operation and target, then verifies page evidence with Playwright.
SOURCE SCREENSHOTEXPAND ↗
An MCP server that lets agents navigate a browser with Jev decisions.
SOURCE SCREENSHOTEXPAND ↗
from the source
Install the Jev Browser MCP server for me. The package is @jkudish/jev-browser on npm an...command is `npx -y @jkudish/jev-browser`; register it as an MCP server with your client....TYPESAFE_API_KEY is already set in the server environment; if not, walk me through setti...pasting the key into the chat (I can create one at console.typesafe.ai/settings/keys). W...registered, ask if I'd like to run a first navigation task, and when we do, show me the...
A browser extension that uses Jev to identify page clutter and save reusable cleanup rules.
SOURCE SCREENSHOTEXPAND ↗
A Codex plugin where Jev handles browser clicks while Codex plans and verifies the larger task.
SOURCE SCREENSHOTEXPAND ↗
from the source
codex plugin marketplace add wy-coliney/jev-browser-use && codex plugin add jev-browser-...
Control macOS by voice or text as Jev navigates the Accessibility tree without screenshots.
SOURCE SCREENSHOTEXPAND ↗
Let Jev choose quick browser actions while a separate language model plans, reads, and drafts answers grounded in page quotes.
SOURCE SCREENSHOTEXPAND ↗
Runs bounded Jev action selection inside Codex Computer Use for observed browser and macOS controls, with local outcome checks.
SOURCE SCREENSHOTEXPAND ↗
A voice-controlled browser where Jev selects an intent and target as speech arrives, then Playwright performs the action.
SOURCE SCREENSHOTEXPAND ↗
from the source
git clone https://github.com/moritzkremb/jev-voice-browser.gitcd jev-voice-browsernpm installnpx playwright install chromiumcp .env.example .env # paste your TypeSafe API key (https://console.typesafe.ai...
Speak a task to your Mac and let local transcription plus one Jev decision choose the operation and target.
SOURCE SCREENSHOTEXPAND ↗
from the source
One TypeSafe request per decision cycle carries:* ``operation`` Choice over the operations that are actually possible on this scree...* ``click_target`` Choice over the observed clickable elements (speculative).The executor consumes only the head that matches the chosen operation.
An experimental Chrome extension that asks Jev whether each DOM element is an ad, then removes matching elements.
SOURCE SCREENSHOTEXPAND ↗
A GUI agent that drives a real computer through USB HID after Jev chooses an operation and indexed screen element.
SOURCE SCREENSHOTEXPAND ↗
from the source
OPERATIONS: "OrderedDict[str, str]" = OrderedDict([ ("CLICK", "Click one element (button, link, tab, checkbox)"), ("TYPE_TEXT", "Type text into one text field or combobox (clears it first)"), ("SELECT", "Choose an option in one dropdown (native select)"), ("KEY_ENTER", "Press Enter to submit the focused field"),
Focus a Windows app, describe a task, and let GhostHand choose accessible controls while risky actions wait for confirmation.
SOURCE SCREENSHOTEXPAND ↗
Give a hiring brief to a visible LinkedIn browser and watch Jev navigate, screen titles, and save profile evidence.
SOURCE SCREENSHOTEXPAND ↗
Generate and replay Excel browser tests with Playwright while local Laya resolves ambiguous controls; Jev support is only a roadmap item.
SOURCE SCREENSHOTEXPAND ↗
A research agent that uses Jev to decide where to browse on Instagram, TikTok and LinkedIn.
SOURCE SCREENSHOTEXPAND ↗
A Chrome extension that uses Jev to check the tone of an X post before it is published.
SOURCE SCREENSHOTEXPAND ↗
Give this native macOS agent a goal and watch it choose guarded clicks and drags from controls detected on your screen.
ACTUAL RECORDING12 SEC ↗
Run a local agent where Jev chooses a tool and its arguments, sees the real result and repeats until it can answer.
SOURCE SCREENSHOTEXPAND ↗
from the source
export type ChoiceQuestion = {type:'choice';instructions:string;criteria:Record<string,s...export type NoulQuestion = {type:'noul';instructions:string;criteria?:{true?:string;fals...export type Question = ChoiceQuestion|NoulQuestion;export type EvaluationRequest = {model:string;state:unknown;questions:Record<string,Ques... const response=await fetcher('https://api.typesafe.ai/v1/systemone',{method:'POST',he...
Mark or fold ads and chosen topics across Xiaohongshu, Weibo, X, and Zhihu as Jev reads each post.
SOURCE SCREENSHOTEXPAND ↗
A browser automation toolkit where a language model plans and Jev chooses each action.
SOURCE SCREENSHOTEXPAND ↗
Match ordinary web-form fields to the right lines in your unstructured personal notes.
SOURCE SCREENSHOTEXPAND ↗
Connect Aside browser agents to Jev for bounded next-action decisions while Aside observes and executes.
SOURCE SCREENSHOTEXPAND ↗
A Mac assistant that enters supplier invoices into Frappe Books, with Jev choosing the on-screen actions.
SOURCE SCREENSHOTEXPAND ↗
from the source
URL = os.environ.get("TYPESAFE_URL", "https://api.typesafe.ai/v1/systemone")def choice(instructions: str, criteria: dict[str, str]) -> dict: return {"type": "choice", "instructions": instructions, "criteria": criteria}def ask(state: dict, questions: dict) -> dict: body = {"model": MODEL, "state": state, "questions": questions}
Turns plain-English browser flows into Playwright runs where Jev selects controls and independent checks preserve evidence for every result.
SOURCE SCREENSHOTEXPAND ↗
Tell wilwid what you don't want to see, and it hides matching posts and videos.
ACTUAL RECORDING12 SEC ↗
Spot the sales pitch, call to action or other intent behind an X post with labels chosen by Jev.
SOURCE SCREENSHOTEXPAND ↗
from the source
content.js 读 DOM、提取帖子、渲染标签、缓存 ↓ chrome.runtime.sendMessagebackground.js (service worker) 唯一发请求的地方 ↓ fetchapi.typesafe.ai
A browser-based 2D autonomous-car simulation driven by Jev decisions.
SOURCE SCREENSHOTEXPAND ↗
from the source
npm installcp .env.example .env # TYPESAFE_API_KEY from https://console.typesafe.ai/settings/k... # optional OPENROUTER_API_KEY for the comparison modenpm start # http://localhost:3000
Drive Playwright from natural-language intents by letting Jev select among the current page’s elements and bounded actions.
SOURCE SCREENSHOTEXPAND ↗
A browser agent with no general-purpose language model: Jev chooses bounded actions, targets and values from each page snapshot.
ACTUAL RECORDING12 SEC ↗
from the source
API_HOST = "api.typesafe.ai"API_PATH = "/v1/systemone"def jev(state, questions, tag): body = json.dumps({"state": state, "model": MODEL, "questions": questions}, ensure_a... headers = {"Authorization": f"Bearer {key}", "Content-Type": "application/json", "Co...
Stamp LinkedIn posts BAIT, CORP, or BRAG as Jev judges them in the feed, with confidence printed beneath each verdict.
SOURCE SCREENSHOTEXPAND ↗
Lets Codex prepare browser goals and text while Jev repeatedly selects the next allowed action, target, and prepared input.
SOURCE SCREENSHOTEXPAND ↗
A browser harness where an LLM plans verifiable subgoals, Jev selects actions and fields, and Playwright executes them.
SOURCE SCREENSHOTEXPAND ↗
from the source
npm start -- --goal 'register a user on example.com' \ --value 'name=Test User' \ --value 'email=user@example.com' \ --value 'password=a-test-password' \ --headed
A browser agent for Ego Lite that asks Jev or another agent to choose one action per observed page state.
SOURCE SCREENSHOTEXPAND ↗
from the source
pnpm exec jev-ego serve --url https://en.wikipedia.org/wiki/Main_Page \ --goal 'Find and open the Wikipedia article about Gödel’s incompleteness theorems.'pnpm exec jev-ego observepnpm exec jev-ego suggestpnpm exec jev-ego act CLICK 4
An MCP browser agent that lets Jev choose each click and map supplied text values to form fields.
SOURCE SCREENSHOTEXPAND ↗
from the source
claude mcp add jev-browse -e TYPESAFE_API_KEY=YOUR_API_KEY -- npx -y -p github:kyrylosyz...
Mark or collapse X replies that resemble generic AI-written paraphrases of the original post.
SOURCE SCREENSHOTEXPAND ↗
Sort loaded social comments into malicious, low-value, inspiring, or ordinary views while surfacing uncertain cases.
SOURCE SCREENSHOTEXPAND ↗
Let jev-ra drive the browser for a coding agent: Jev picks the next action and the page element to use.
ACTUAL RECORDING12 SEC ↗
A Chrome extension that uses Jev to navigate websites while guarding irreversible clicks.
SOURCE SCREENSHOTEXPAND ↗
A Go command-line tool that uses Jev to control macOS applications.
SOURCE SCREENSHOTEXPAND ↗
from the source
computer-use-jev -goal 'in TextEdit, type "hello world"'computer-use-jev -goal 'open Safari and take a screenshot' -jsoncomputer-use-jev -goal '...' -dry-run # decide, don't actcomputer-use-jev -goal '...' -max-steps 24
Browser automation with Jev choosing what to click and Playwright doing the clicking.
SOURCE SCREENSHOTEXPAND ↗
from the source
npx jev-browser open https://example.com --session worknpx jev-browser snapshot --session work# Use the ref returned by snapshot, or a caller-authored Playwright selector:npx jev-browser click 'a' --session worknpx jev-browser take_screenshot --args '{"filename":"page.png","fullPage":true}' --sessi...
A frontend testing project built on Jev Ultrafast and Browser Harness, with evidence capture and a synthetic todo demo.
SOURCE SCREENSHOTEXPAND ↗
A privacy-focused Chrome extension that uses Jev to identify native ads, sponsored feed cards, and video ads.
SOURCE SCREENSHOTEXPAND ↗
from the source
# Clone the repositorygit clone https://github.com/vmendes90/jev-shield.gitcd jev-shield# Install dependenciesnpm install
Search an article by concept and highlight matching words even when the query itself never appears.
SOURCE SCREENSHOTEXPAND ↗
Grade every visible page section for clarity, writing quality, and on-page SEO from a browser extension.
SOURCE SCREENSHOTEXPAND ↗
Watch suspected ads crumble into ash after Jev classifies their compact DOM descriptions.
SOURCE SCREENSHOTEXPAND ↗
Detect sponsored segments from Bilibili subtitles and skip them when Jev’s independent ad and product judgments clear a confidence threshold.
SOURCE SCREENSHOTEXPAND ↗
Blur posts Jev classifies as likely AI-written, with an uncertain result available when the text does not support a confident distinction.
SOURCE SCREENSHOTEXPAND ↗
Watch a browser agent split quick target, completion, and danger judgments from slower planning in a live control panel.
SOURCE SCREENSHOTEXPAND ↗
Almond's Chrome DevTools browser agent uses Jev for fast action selection and includes the Browser Use Olympics benchmark.
SOURCE SCREENSHOTEXPAND ↗
from the source
open -na "Google Chrome" --args --remote-debugging-port=9333 --user-data-dir=/tmp/fastlo...echo 'TYPESAFE_API_KEY=...' > ~/.config/typesafe/env # chmod 600node olympics.mjs # full course, no planner; registers as "Almond-fastloop"BUO_PLANNER=1 node olympics.mjs # same course with a Sonnet planning call before the c...node almond-fastloop.mjs task_booking.json # single task
A Chrome extension that uses Jev to analyze X posts as they appear in the timeline.
SOURCE SCREENSHOTEXPAND ↗
Automate a Mac or Windows desktop with Otto, which reads the screen locally and asks Jev what to do next.
SOURCE SCREENSHOTEXPAND ↗
Jev, briefly
A decision model, not a chatbot.
A chat model generates text token by token and hopes you parse it. Jev never generates a word. Your code sends the state of the world plus typed questions; Jev returns every answer in one parallel pass — typed values with calibrated probabilities and a confidence score your code can trust. About 70–500 ms end to end, $0.042 per million input tokens, output free. Trained with what TypeSafe calls Reinforcement Learning for Calibrated Decisions.
choice Pick one of these.
You define the options. Jev returns the chosen option, a probability for every option, and a confidence. The workhorse of routing, agents, and games — the answer space is always legal, so there is nothing to parse and nothing to hallucinate.
score Rate this on a rubric.
You describe ordered levels — say trivial / normal / critical. Jev returns the level, the probability of each, and a confidence. Scores turn fuzzy judgment (“how severe is this log line?”) into a number-free decision your code can branch on.
noul Is this statement true?
You assert a statement; Jev returns the probability that it is true — a noul. Moderation, verification, guardrails, “does this diff actually fix the bug?”: one question, one calibrated probability.
Official material lives at typesafe.ai and docs.typesafe.ai. JevMade is an independent community registry — not affiliated with or endorsed by TypeSafe AI.
Recurring lessons
Patterns that keep showing up.
One decision per tick
Games, drones, trading bots, and browser agents all converge on the same loop: serialize the state, ask one decisive question, act, repeat. Jev’s latency makes the loop feel instant — the model lives inside the control loop, not outside it.
Confidence as a gate
The answer says what; the confidence says whether to act. The most reliable entries threshold on confidence to route edge cases to a slower model or a human — automation with an honest escape hatch.
Shrink the choice space in code
Makers rarely ask Jev to pick from everything. Local tactics prune 225 gomoku moves to ~40; DOM filters turn a page into an element table; code narrows options, Jev judges within them.
Jev judges, LLMs talk
A recurring split: Jev makes every decision cheaply and instantly, and a small LLM is only invoked when a human-facing string must actually be written. Decision and generation are separate budgets.
Made something with Jev?
Send the link — repo, live demo, post, or video — plus a line on what it does and which primitives it uses. Every entry is verified against its primary source before it ships.
Submit your experimentSign in first so we can reply to you. Submissions are reviewed, not published automatically.