Submit a project
Choose by the job, then check the boundary
- Pick the integration surface. For an existing Playwright MCP client, inspect a proxy first. For direct browser scripts, compare a CLI/SDK. For reproducible testing, start with a test runner and explicit assertions.
- Trace observation to action. Read what each project sends to Jev, which controls it permits, and what executes the selected action. Test a missing or disabled control and an untrusted page instruction.
- Define success outside the model. Check the page state after each action. For a saved record or payment, require application evidence beyond a model's “done” response or a UI readback.
- Check adoption costs. Verify runtime, browser, API key, data exposure and license in the linked README before installing. The entries below summarize documentation; Jev Atlas has not run or security-audited these projects.
Eight starting points
| Project | Start here when | Check before use | Source |
|---|---|---|---|
| Jev Ultrafast | A Chrome browser loop that chooses an operation and visible control for a narrow goal. | Requires uv, your own TypeSafe and text-model keys, and Chrome through Browser Harness; a DONE choice still needs an independent outcome check. | README |
| Mobile Jev | An Android device agent with a local studio and a dark-theme demo that checks the resulting switch. | Needs a Mobilerun device, Node/pnpm and separate Mobilerun and TypeSafe keys; public hosting needs your own access controls. | README |
| Computer Use Jev | A native macOS accessibility agent for already running apps, such as a TextEdit typing task. | macOS 14+, Go, Xcode Command Line Tools, Accessibility permission and a TypeSafe key; it does not launch apps or inspect browser DOM. | README |
| Jev Browser by tontoko | A Playwright CLI, SDK and MCP server with native assertions. | Node 22.15+; check GitHub release installation. UI readback does not prove a database write. | README |
| Jev Browser by KesavanKing | A local visual panel for stepping through observed targets and exporting action traces. | Requires Python 3.12+, uv and a separate text model for field values. | README |
| Jev Playwright MCP | A proxy that adds page triage and action checks to an existing Playwright MCP setup. | Requires Node 20+ and Chromium; model-based masking and gates are fallible, so keep explicit tool policy. | README |
| JevTest | Isolated Playwright flows with deterministic assertions and replayable traces. | Requires Node 24 and pnpm 11. Its README says usage licensing is pending. | README |
| Jev Browser by MahmoudAdelbghany | An MCP browser loop that returns control to the calling agent when a step is uncertain or stuck. | Requires Node 20+, Chromium and a TypeSafe key. Its README says it is not a security boundary. | README |
Check the source date on each project page; requirements can change.
Try a bounded example
Our browser-control experiment documents its synthetic pages, action checks and timing limits. The question-design guide helps you define a “none” option before automating a click. For searching web evidence rather than controlling a form, compare the psearch project with the semantic reranking task guide.
Browser-control experiment · Question-design guide · psearch · Semantic reranking guide