User repository on GitHub
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
Jev Browser — vinilana
TypeScript browser harness where an OpenRouter planner splits goals into subgoals, Jev picks each concrete action, OpenRouter fills text fields, and Playwright drives the page.
- Stars
- 4
- Language
- TypeScript
- Published by
- Community
- Author
- vinilana
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is Jev Browser — jkudish.
About this project
A hybrid browser usage harness in TypeScript: an OpenRouter LLM turns goals into verifiable subgoals, Jev chooses each action and DOM field, OpenRouter generates text whenever a field needs filling, and Playwright executes the actions. Create a .env file using .env.example and set TYPESAFEAPIKEY. Get your key from the TypeSafe console. The default model is jev-latest; set TYPESAFEMODEL to change it.
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Goal mode plans verifiable subgoals while direct commands and batches skip planning and run in order.
- Jev chooses from concrete observed actions with a confidence gate; fills route through OpenRouter with missing-input stops.
- Deterministic navigation checks, automatic waits, and repeat detection bound loops before the step limit.
- Every run writes observations, screenshots, traces, and a separate Jev evaluation with explicit exit codes.
Watch out
Browser behavior and site structure vary widely, so use test accounts and expect loop, confidence, and evaluation handling on unfamiliar pages. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
Planning needs prose generation, action choice needs calibrated selection, and filling needs text. Jev observes the page and picks one concrete action at a time while the planner and generator handle language tasks.
Labeled values guide the generator, but missing data returns null and stops with a missing-input status instead of inventing values. Password values are redacted from logs, so use test environments for real effects.
Step caps, a five-wait limit, low-confidence stops, repeat detection, and explicit incomplete statuses bound the loop. Headed mode, observations, and per-step artifacts make stalls inspectable.
Only partly. Completion is an executor claim checked by a separate Jev evaluation over commands, final state, and concrete actions. Offline reports bundle actions, confidence, timing, evidence, and the verdict.
Use Node.js 22 or later, run npm install and npm run browser:install for Playwright, then create a .env from .env.example and set TYPESAFE_API_KEY. The default model is jev-latest and TYPESAFE_MODEL changes it; field filling also needs OPENROUTER_API_KEY.
Only fills go through OpenRouter, and every fill does even when you supply the value. Clicks, Enter, selections and scrolling do not call it, and standalone navigation completes deterministically from the observed URL, recorded as decisionSource postcondition with confidence 1.
Each run writes to runs/<timestamp>/: report.html, plan.json in goal mode, events.jsonl, steps.jsonl, per-step before and after observation JSON and screenshots, a final observation, a Playwright trace.zip, eval.json and result.json.
More Jev resources like this
7GitHub stars
Demo that turns free-text UI descriptions into real shadcn components using only Jev typed Choice, Noul, and Score answers, with no generated code on the decision path.
Jev Demos & Experiments#community#jevbest#catalog-import
7GitHub stars
Two TypeScript monorepo products that keep the rules in the engine and ask Jev exactly one question: what should this person or this shogi side do next. Hundred runs a hundred NPCs with hunger, money, jobs, relationships and memories; Jev Shogi picks Gote's reply from engine-legal USI moves.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
jammaru/jev-labTypeScript
jammaru/jev-labTypeScript4GitHub stars
Open-source Electron computer-use agent for macOS and Windows that reads the apps you select through native accessibility with local OCR added for text targets, and asks Jev to choose concrete actions that it shows you before executing.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
NobleSpartan6/ottoTypeScript
NobleSpartan6/ottoTypeScript1GitHub stars
A decision board that sends one System One request describing eleven stream ideas, then lets code gate and weight Jev's answers into a single pick for what to build next.
Jev Demos & Experiments#community#jevbest#catalog-import
0GitHub stars
LaneBreak is a support ticket triage demo covering priority assignment plus routing, built on TypeSafe Jev. The README is a title, one sentence, and the standard npm i and npm run dev pair.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
ndolinschi/lanebreakTypeScript
ndolinschi/lanebreakTypeScript0GitHub stars
SwarmRouter picks which agent or skill should handle a task and returns a confidence with the pick, drawn as a visual swarm map, with the choice made by TypeSafe Jev. Five handler types are named in the README.
Jev Demos & Experiments#community#jevbest#catalog-import
Communityndolinschi/swarmrouterTypeScript
ndolinschi/swarmrouterTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.