Jevry
Open-source desktop browser agent where a language model plans, Jev chooses the executable action and Chromium carries it out.
Overview
Jevry is an open-source desktop browser that works through supported websites from plain-language tasks, with visible actions, page-grounded results and controls to stop or redirect. Its split is 'language models plan, Jev decides, Chromium acts': Jev chooses among actions the browser can actually execute, a separate text model handles planning, language and occasional visual reasoning, and the desktop runtime validates and executes actions and retains evidence.
Tabs and conversations live in one workspace, so you can ask it to find documentation, compare sources or operate page controls and then check the result. The README shows a recorded run in which it plays 2048 to a win (20,708 points in 1,013 moves) and has a benchmarks section; it notes that website support varies. It is a source preview requiring Node.js 22.12+, a TypeSafe key and a separate text-model connection, and model usage costs money.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which executable browser action should be taken next? | — |
How it works
- User gives a task in plain language
- Text model plans the next sub-goal
- Jev picks among actions the browser can execute
- Runtime validates and executes the action in Chromium and records evidence
- User can stop or redirect at any time
Strengths
- Actions are visible and grounded in the page
- Jev limited to executable actions
- Evidence retained for checking results
Caveats
- Source preview, not a packaged release
- Needs both a Jev key and a text-model connection
- Website support varies
Getting started
See the Install section and docs/FIRST_SUCCESS.md; requires Node.js 22.12+, a TypeSafe key and a text-model connection.At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- desktop app (Chromium)
- Has code
- yes
Use cases and integrations
How curated lists describe it
Desktop browser agent where a language model plans, a decision model picks the action, and Chromium carries it out.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
Which executable browser action to take next; planning and language come from a separate model.
Is it ready for everyday use?
It is a source preview and website support varies.
What does it cost?
You pay for Jev and the text model through your own keys.
Related projects
Lets coding agents run whole goals in your signed-in Chrome on macOS in background tabs, asking Jev or Claude Haiku only for real choices.
ProjectbetaComputer-use platform (Cua Spaces, Driver, Lume) with CUA-S1 small decision models and a preview jev-use example for bounded actions.
ProjectbetaPythonLocal stdio MCP server that executes whole browser and macOS goals with a Jev-driven observe-decide-act loop.
MCP serverbetaTypeScriptBrowser Use's browser agent where Jev picks an operation and page element in one request and a small LLM only writes text to type.
LibrarybetaPythonPython decision layer for browser and desktop automation: code observes, acts and verifies; the model makes one typed choice per step.
ProjectbetaPythonBrowser and computer use for coding agents via CLI and MCP, with Jev's calibrated judgments and a code-owned control loop.
MCP serverbetaTypeScript