User repository on GitHub
Evidence-driven frontend QA built on Jev Ultrafast and Browser Harness, with a synthetic todo demo.
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
Jev Frontend Qa
Python 3.12 CLI that drives real Chrome through Browser Harness and asks Jev to pick each next browser operation and target, while scenario fixtures, deterministic assertions and fresh persistence reads decide the actual contract verdict.
- Stars
- 5
- Language
- Python
- License
- MIT
- Published by
- Community
- Author
- Nainish-Rai
- Last verified
- Oct 2026
About this project
Evidence-driven frontend QA built on Jev Ultrafast and Browser Harness. Python 3.12+, Chrome/Chromium, plain HTML/CSS/JavaScript demo, and SQLite.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- Jev is only the action selector: it never authors the scenario, generates selectors or JavaScript, or invents field values.
- One POST to the System One endpoint per decision, batching an operation question with target questions; the default action threshold is 0.55.
- Low confidence stops the run as BLOCKED, and Jev's DONE means stop interacting rather than the test passing.
- Installable with uv tool install and shippable as a bundled jevqa skill for Claude Code or Codex.
Watch out
Jev receives redacted structured page observations, not screenshot pixels; screenshots are written locally and are unredacted, so they need explicit policy permission. A TypeSafe key is required for browser runs but not for schema, init or validation commands. The CLI validates and authorizes itself and starts a private browser session unless attaching was authorized; it never starts your app server. init writes jevqa/policy.json with no permitted origins and no model disclosure, so an unchanged policy intentionally blocks a live run, and installing the skill is not that approval. There is no offline-model fallback in the production CLI. Planned exploration additionally requires allow_page_text and allow_planner, the latter defaulting to false, and runs the host CLI with tools disabled. 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.
No. Jev's DONE means stop interacting, not that the contract passed. Authored assertions check the actual DOM and correlated HTTP evidence, and persistence assertions perform fresh authorized reads. In the Todo example a success toast with no saved record fails the contract even though the model thinks the interaction finished.
Jev gets permitted, redacted structured page observations plus the step goal and supplied fixtures, and up to ten recent actions if action-history disclosure is on. Captured HTTP request and response bodies are local assertion evidence and are not sent. Screenshots stay local, are unredacted and require explicit policy permission.
jev-qa init writes jevqa/policy.json with no permitted origins and no model disclosure, on purpose. You have to approve your app's origins, HTTP methods and path prefixes, and the permitted synthetic observations, and keep the scenario start_url consistent with that policy before a live run will proceed.
No. explore lets an explicitly selected, authenticated Claude or Codex CLI choose bounded subgoals while Jev remains the browser-action selector and the local runner authorizes every operation. It needs both model_disclosure.allow_page_text and allow_planner, the latter false by default, and the host runs in a temporary directory with its tools disabled. Exact fixture values and correctness checks stay in code either way, so a wrong model answer becomes a BLOCKED run rather than an arbitrary click.
Python 3.12 or newer, uv, Chrome or Chromium, and a TypeSafe API key for live runs. Schema, initialization, scenario registration, and validation commands need neither a key nor a running browser.
The default action threshold is 0.55. Code accepts the answer only if it is a valid offered choice, and low confidence stops the run as BLOCKED rather than guessing.
Duration, actions, assertions, screenshots, Jev calls, latency, tokens, and estimated cost, shown in the terminal and written to the JSON report.
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-forgePython70GitHub 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/stuntdPython42GitHub stars
Open harness that renders allowed JSON answers on a diffusion canvas, denoises for a fixed step budget, then picks the top allowed tokens from final logits with no training.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
JoshuaSP/open-jevPython
JoshuaSP/open-jevPythonFrom 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.