ego-jev
Agent skill porting jev-ultrafast's loop into ego lite: one ~0.4 s Jev call per DOM step picks operation and target.
Overview
ego-jev ports the inner loop of browser-use's jev-ultrafast into ego lite, so an agent keeps your real browser, sessions and logins while each page step costs one fast typed decision instead of a full LLM turn. Ego executes and Jev decides: each step snapshots the page, numbers the interactive elements, and one System One call picks both the operation and its target, then ego clicks, fills or selects.
The README reports about 300 to 550 ms and roughly $0.0005 per decision, and on a seven-step form task a median Jev decision of about 270 ms versus about 1 to 2.8 seconds for a bare LLM call over the same snapshot. It also finds clickable elements the accessibility snapshot misses (div cards, pointer widgets, iframe and shadow content), and escalates rather than guessing on payments, deletes, uploads, confirmations, logins, free text, low confidence and repeated actions. Done is treated as a claim that your verify step decides.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which operation on which numbered element should be performed next? | — |
How it works
- Snapshot the page and number interactive elements
- One Jev call picks operation and target
- ego executes the action
- Escalate to the user on risky actions, free text or low confidence
Strengths
- Uses your real browser and logins
- Fast, cheap per-step decisions
- Escalation on sensitive actions
Caveats
- Requires ego lite
- Page content sent to TypeSafe
Getting started
npx skills add ZephyrDeng/ego-jev (or install as a Claude Code plugin).At a glance
- Type
- Agent skill
- Status
- beta
- Language
- TypeScript
- Runtime
- agent skill for ego lite
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Jev (TypeSafe System One) inner loop for ego-browser — one ~0.4s typed decision per DOM step instead of an LLM turn. Agent skill for ego lite.
logicrw/awesome-jev-projectsJev (TypeSafe System One) inner loop for ego-browser — one ~0.4s typed decision per DOM step instead of an LLM turn. Agent skill for ego lite.
hellogumbo/awesome-jevAgent skill that decides each DOM step from the page's candidate elements rather than spending an LLM turn on it.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
The next operation and its target element on the page.
How fast is a step?
About 0.3 to 0.55 seconds per decision per the README.
What happens on risky actions?
Control is handed back to you.
Related projects
MCP server and libraries that drive Chrome over CDP and hand a decision model like Jev a bounded menu of page actions.
ProjectproductionTypeScriptAgent-plugin browser tool where Jev picks one operation and one row from a numbered page table; a small LLM only writes typed text.
ProjectbetaTypeScriptChrome extension plus MCP server so Claude or Codex can drive your real Chrome, with Jev picking each click in about half a second.
MCP serverbetaTypeScriptJev browser agent that pierces shadow DOM and iframes, adds canvas fallback and full actions, with one Jev request per cycle.
MCP serverbetaPythonMCP server that lets Claude Code or Codex drive real Chrome with Jev picking elements and checking steps, plus safety gates and traces.
MCP serverbetaTypeScriptFast browser-use layer for coding agents: Jev picks the operation and target element each step, via MCP server or CLI.
MCP serverbetaPython