User repository on GitHub
Pure Jev that can "type" and drive towards task completion.
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
JevOnly
Browser automation loop with no LLM where code enumerates page actions and Jev only picks, verifies, and stops, including a live viewer and QA regression mode.
- Stars
- 5
- Language
- Python
- License
- APACHE-2.0
- Published by
- Community
- Author
- buluoray
- Last verified
- Oct 2026
About this project
Pure Jev that can "type" and drive towards task completion. No LLM anywhere: code builds every option from the page and the goal, Jev only picks — and that is enough to fill forms, type, and finish the task. Needs Python 3.12+, Node.js 22+ and a TypeSafe API key (tsk…). run.sh creates a venv, installs Chromium, and opens the live viewer at paste the key, a start URL and a goal, press Run, and watch every observation, vote, action, verification, copy and undo as it happens.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- Observes enabled controls and visible text, votes on next action, binds only copied values.
- Acts through Chromium then verifies before and after, undoing failures automatically.
- Gates risky actions by risk vote with refuse, ask, and allow policies per surface.
- Reuses the loop as a QA suite runner with pass tables, reports, and replayable logs.
Watch out
Goals, facts, and page observations are sent to TypeSafe, and stopping without a code-owned check rests on Jev judgement. 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.
Code builds every option from the page, the goal, and copied values, then asks Jev closed questions for done status, next action, and verification. Fields only receive goal spans or page-copied values, never invented text.
The Wikipedia goal finishes in about 20 steps with dozens of Jev calls in half a minute, and a QA case costs about a cent per run. Irreversible actions above the risk vote go through a refuse, ask, or allow gate.
Paste the key, start URL, and goal into the local viewer and watch each observation, vote, action, and undo live. CLI mode takes start URL, goal variant, and output path instead.
Point the same loop at a site you own with expected text, URL, selected control state, or answer substrings as code-owned checks. It prints pass tables, keeps event logs for replay, and exits non-zero on failure.
The goal, the facts and the page observations go to TypeSafe, so point it at pages you are willing to send there. The viewer binds to 127.0.0.1 only, the key stays in the tab's sessionStorage and the local process, and screenshots and replays never leave the machine. Nothing irreversible runs by default, and a site that blocks automation is treated as a wall rather than bypassed.
Five moves. Observe the accessibility snapshot of enabled controls and visible text. Vote in one request on done?, off path? and which action next. Bind values, so a field can only receive a fact, a span of the goal, or a value copied from the page. Act in Chromium and let Jev compare before and after, undoing failures. Then stop when Jev says so, when every value the goal asked for is in hand, or when a code-owned check passes.
The rules live in jevonly.core and are environment-agnostic; the browser in jevonly.envs.browser is only the first environment, and docs/architecture.md documents writing another. Development is make install, make test and make lint, and the code is Apache-2.0. Two limits are stated plainly: goal splitting is English-centric, and the viewer runs one run at a time.
More Jev resources like this
46GitHub stars
Open 150M encoder that reproduces the typed noul, choice, and score interface in one forward pass, trainable on a free GPU and runnable locally for nothing.
Jev Demos & Experiments#community#jevbest#catalog-import
941GitHub stars
Library that adds schema-guaranteed typed outputs to existing autoregressive LLMs without changing weights, served through SGLang or a hosted API.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
TypeLLM/TypeLLMPython
TypeLLM/TypeLLMPython55GitHub stars
Toolkit that makes a model behind an OpenAI-compatible endpoint or running in-process answer typed choice, score, and yes/no questions from next-token probabilities in Jev API shape.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
fidecastro/jevifyPython
fidecastro/jevifyPython164GitHub stars
End-to-end toolkit that synthesizes decision data, trains and calibrates candidate scorers, evaluates fixed splits, and serves Jev-compatible inference for interactive web decisions.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
zwliJay/jev-forgePython
zwliJay/jev-forgePython63GitHub stars
Claude Code kit that guards PreToolUse calls with a free pre-filter plus one Jev question for ambiguous cases, and adds sub-agent sizing, rechecks, search, daemon speedups, and monitoring.
Jev Demos & Experiments#community#awesomejev#catalog-import
Community
jonathanavis96/jev-kitPython
jonathanavis96/jev-kitPython70GitHub stars
Self-hosted proxy that records typed decisions, distils them into small heads on a frozen encoder, and answers locally when confident.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
bladedevoff/stuntdPython
bladedevoff/stuntdPythonFrom 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.