What is jev-browser-local?

A fully local fork of jkudish/jev-browser: the browser CLI's every judgment and every typed string is computed on your own machine through the parallel-decisions engine, with no cloud API and $0 per decision.

How does it avoid calling TypeSafe or OpenRouter?

A single Python bridge, local_jev_server.py, implements TypeSafe's POST /v1/systemone contract on top of the local parallel-decisions engine and points jev-browser at localhost. The hosted Jev transport is never used.

What models does it run?

Qwen2.5-1.5B-Instruct makes the decisions and Qwen2.5-0.5B-Instruct does the typing, meaning generating search queries, as parallel constrained decoding on CUDA with torch, fp16 and no CUDA graphs.

What hardware do I need?

The README's setup runs on an RTX 2060 SUPER with a CUDA venv containing parallel-decisions[torch]. Timing figures such as 5.9 seconds for the Wikipedia run come from that one machine and will vary elsewhere.

How fast was the recorded run?

The 2026-09-18 run sent jev-browser to Wikipedia's main page and stopped on the Ristretto article: status goal_achieved, 3 steps, 5.9 seconds, roughly zero dollars, with JSON traces and screenshots saved under runs/.

Is it safe to install?

The README publishes a pre-install security review: no telemetry or phone-home, a postinstall script that only runs npx playwright install chromium (opt out with JEV_BROWSER_SKIP_BROWSER_DOWNLOAD=1), and model output that cannot become code because the model picks an ID from a fixed list of observed elements.

Is this an official TypeSafe project?

No. The README states it is an independent project not affiliated with TypeSafe AI, based on jkudish/jev-browser v0.4.0 (MIT) with a warm-browser/CDP fork, and the repository's own license shows as NOASSERTION on GitHub.

Jev directory

CommunityJev Repos & SDKs

Jev Browser Local

A fork and bridge that runs jkudish/jev-browser fully locally: a local Jev-style engine built on parallel-decisions replaces the hosted API, with Qwen2.5-1.5B-Instruct picking each browser step and a 0.5B model typing search queries, measured on an RTX 2060 SUPER.

See highlights Open github.com

Stars: 3
Language: Python
Reading time: 2 min
Published by: Community
Author: rorshopping
Last verified: Oct 2026

rorshopping/jev-browser-local

User repository on GitHub

View on GitHub

Run jev-browser on a fully local JEV-style decision engine (no cloud API). Warm-browser fork, VRAM guard, measured benchmarks, run traces.

Forks: 1
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity catalog-import logicrw local-inference browser-automation cuda

Works forJev for Realtime control Jev for Tool calling

Used inJev for Software & DevTools

Languages: Python JavaScript
Platforms: CLI MCP Browser Playwright
Capabilities: Open sourceRuns locallySelf-hosted

Highlights

Replaces every TypeSafe API call with a local engine: no TypeSafe API, no OpenRouter, no OpenAI, $0 per decision.
Final configuration uses Qwen2.5-1.5B-Instruct for decisions and Qwen2.5-0.5B-Instruct for typing search queries.
A measured Wikipedia run reached the Ristretto article in 3 steps and 5.9 seconds with goal_achieved status in run-success.json.
local_jev_server.py speaks TypeSafe's POST /v1/systemone contract and adds an OpenAI-compatible /v1/chat/completions typing endpoint.
A pre-install security review found no telemetry, and actions resolve from a fixed list of observed elements stamped with data-jev-id, never raw selectors.

Quickstart

bash

jev-browser run "Search Wikipedia for the espresso-based drink called Ristretto and stop when you are on that article" https://en.wikipedia.org/wiki/Main_Page

Watch out

Grounded in the repository README and GitHub metadata reviewed on 2026-10-15; all timing numbers come from a single RTX 2060 SUPER machine, the install commands use development-machine absolute paths and must be adjusted, the fork is based on jev-browser v0.4.0 (MIT) while the repo's own license shows NOASSERTION, and the project is independent of TypeSafe AI.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

The Open Reimplementations: Local Models That Speak System One

Sep 22, 2026

Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.

Jev Repos & SDKs#open-models#research#local

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai