jev-browse
github.com/danielnc/jev-browse- Category
- Developer Tools
- Rank
- No. 1732Tools index
Previous survey · No. 1710 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Workflow Automation
- GitHub
- 10 stars
- Latest release
- v0.3.0
- Date
About
jev-browse is an open-source library that adds typed browser sub-task helpers on top of browser-harness, letting coding agents like Claude Code or Codex delegate multi-step website tasks (e.g., search, filter, open a result) in a single call. Each step is resolved by TypeSafe's Jev, a small model that returns a typed decision in under a second, so the calling agent gets a structured outcome or a typed reason for stopping instead of raw browser output.
At a glance
jev-browse lets a coding agent hand a bounded browser task to a helper that uses Jev to choose steps. It is worth evaluating for repetitive navigation or form work when several full agent turns are slowing it down.
Good fit
Developers using browser-harness who can give a task an observable stopping condition.
Consider something else when
Keep a human in control when a task uses real accounts, uploads, payment or irreversible actions.
Start here
Try a read-only navigation task in a disposable browser. Verify the final URL or selected result yourself and compare the same task with direct agent control.
Cost and requirements
MIT-licensed. The documented setup needs Chrome, browser-harness, Python 3.11 or newer and a TypeSafe API key.
Limits to know
The service helps choose browser actions and can return a stop reason. Check what browser context leaves the machine before using private pages. The author’s timing and cost comparisons are workload-specific.
Sources checked 2026-11-05. Source-based assessment.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.