Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

Chrome Relay

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.

ProjectbetaMIT
View on GitHub ↗Live demo ↗

Overview

Chrome Relay lets coding agents such as Claude Code, Codex, Cursor and Paseo use your everyday Chrome on macOS, with your real logins, from the command line in background tabs that never take focus. Agents can hand over a whole goal in one command, such as inviting a teammate as an admin with given data and an until condition, and Chrome Relay carries it out and checks each step. Step-by-step commands also work at about 0.15 s each.

For each step it fills what the agent named, replays flows that worked before on that site, asks a fast decision model only for real choices, and hands back anything it is unsure of. The decider is TypeSafe's Jev (about 0.15 s a step) if you saved a key, else Claude Haiku through your own Claude Code (about 1 to 1.5 s). The README reports a 7-step form in 2.5 s with Jev and 1.2 s replayed. With Jev on, page text from agent tabs goes to TypeSafe. It is opt-in and Mac-only.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich control should be used for this step of the goal?(inferred)controls on the page

How it works

  1. Agent issues chrome-relay do with a goal and data
  2. Fill named fields and replay saved flows
  3. Ask Jev (or Haiku) for remaining choices
  4. Check each step against the until condition
  5. Hand back uncertain steps to the agent

Strengths

  • Uses real logins without stealing focus
  • Model called only for genuine choices
  • Fast per-step decisions with Jev

Caveats

  • macOS only
  • Page text goes to TypeSafe when Jev is on

Getting started

See https://agent-chrome-relay.aindeev.com for setup; save a key with pbpaste | chrome-relay decider jev.

At a glance

Type
Project
Status
beta
Runtime
Chrome extension + CLI (macOS)
Has code
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Lets coding agents use your signed-in Chrome on macOS in background tabs, carrying out a whole goal per command with Jev (or Claude Haiku through Claude Code) choosing only the steps it cannot fill or replay. With Jev on, page text of agent tabs goes to TypeSafe.

    MrJev/awesome-jev

Frequently asked questions

When is Jev used?

Only for steps that cannot be filled from the agent's data or replayed.

What if I have no Jev key?

It uses Claude Haiku through your own Claude Code.

Does it take over my screen?

No, it works in background tabs.

Related projects

  • system1-agents

    Prebuilt browser, computer-use, robotics and game agents whose actions come from Jev, Laya or Cua-S1, with a scaffold for new agents.

    ProjectbetaPython
  • jev-desktop

    Codex plugin adding a bounded Jev decision loop to Computer Use for browser tabs and native macOS apps.

    Projectbeta
  • Jev GUI Delegate

    Windows and Chrome GUI delegation for Codex: a local controller runs a task contract, calling Jev only for semantic choices.

    ProjectprototypePython
  • public-browser

    MCP server and libraries that drive Chrome over CDP and hand a decision model like Jev a bounded menu of page actions.

    ProjectproductionTypeScript
  • cua

    Computer-use platform (Cua Spaces, Driver, Lume) with CUA-S1 small decision models and a preview jev-use example for bounded actions.

    ProjectbetaPython
  • Flick

    Local stdio MCP server that executes whole browser and macOS goals with a Jev-driven observe-decide-act loop.

    MCP serverbetaTypeScript