jev-agent-browser
Bounded browser execution for parent agents: Jev picks the next typed action, agent-browser runs it, stuck states hand back.
Overview
jev-agent-browser gives a parent agent a way to delegate bounded browser tasks. Given a URL, a goal and a step limit, Jev chooses the next typed action, Vercel's agent-browser executes it, and ambiguity, repetition or a stuck state becomes a structured handoff back to the parent. A parent can also supply semantic goal values, such as a destination URL, without exposing DOM references.
The README shows a 17-second read-only walkthrough on the Hugging Face models page. Configuration in ~/.config/jev/config.json sets the decision endpoint (for example OpenRouter's decisions API), the environment variable holding the key, the model, pacing between actions and maximum steps; configured tools are offered to Jev as bounded RUN_TOOL choices. API keys stay in environment variables and never enter page or decision state. It is published on npm and MIT licensed, and also offers research and classification tools.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What is the next typed browser action toward the goal? | — |
| Yes / no | Is the task ambiguous, repeating or stuck and should be handed back?(inferred) | yes · no |
How it works
- Parent agent calls jev with URL, goal and max steps
- agent-browser snapshots the page
- Jev picks a typed action (or RUN_TOOL)
- agent-browser executes; repeat within bounds
- Return success or a structured handoff
Strengths
- Bounded steps keep runs predictable
- Structured handoff instead of silent failure
- Keys never enter page or decision state
Caveats
- Depends on agent-browser as a separate tool
- Needs a TypeSafe or OpenRouter key
Getting started
`npm install -g agent-browser jev-agent-browser`, set TYPESAFE_API_KEY, then `jev --url https://huggingface.co/models --goal 'Open Tasks and select Text Classification' --max-steps 6`.At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- Node CLI with agent-browser
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Browser agents: a parent agent delegates bounded tasks to a Jev loop that selects typed browser actions, validates them through agent-browser, and escalates ambiguity or stuck states back to the parent.
yibie/awesome-jevFast, bounded browser agents powered by Jev and agent-browser — typed actions, research, classification, and safe orchestration.
logicrw/awesome-jev-projectsFast, bounded browser agents powered by Jev and agent-browser — typed actions, research, classification, and safe orchestration.
hellogumbo/awesome-jevBounded browser tasks for a parent agent: Jev picks the next typed action, agent-browser runs it, and ambiguity or a stuck state comes back as a structured handoff.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
The next typed browser action at each step, within a bounded loop.
What happens when it gets stuck?
It returns a structured handoff to the parent agent.
Can I use OpenRouter?
Yes, set the decision endpoint and key variable in the config file.
Related projects
Chrome extension port of Jev Ultrafast that drives your current tab, with Jev picking the operation and element each step.
Browser extensionbetaTypeScriptPython decision layer for browser and desktop automation: code observes, acts and verifies; the model makes one typed choice per step.
ProjectbetaPythonManifest V3 Chrome port of jev-ultrafast that drives your current tab, with Jev choosing each operation and element.
Browser extensionbetaTypeScriptPython browser agent where Jev picks each operation and element in about 150 ms, escalating to an LLM only when unsure.
ProjectprototypePythonReference framework where Jev picks a concrete computer-use action and its execution route from legal, verifiable candidates.
AppresearchPythonAutoGPT agent platform, listed as adding Jev decision blocks for typed routing, filtering, scoring and confidence-gated next actions.
ProjectproductionPython