Open-source browser extension

AgentBridge

Human approval for AI agents.

Add to Chrome View source on GitHubMIT license

Overview

A checkpoint between agent and browser.

AgentBridge adds a human approval step between AI agents and your browser. An agent can read the active tab, use the clipboard, and type into a page - but every action pauses until you approve it. Free, no account required, and it runs entirely on your machine.

The AgentBridge side panel asking to write the clipboard, with approve and deny buttons — The agent asks. Nothing runs until you approve it.

What it does

A pause before every action.

Letting an agent operate your browser unattended is a bad idea - it can click, paste, fill forms, or copy sensitive data faster than you can react.

Approval prompts

A clear dialog before every agent action, showing exactly what it wants to do.

Activity log

A full history of every action the agent requested and what you decided.

Clipboard control

The agent can read and write the clipboard, but only with your permission each time.

Page interaction

The agent can type into a page - form filling, autocomplete - only after you approve.

How it works

Every request waits for you.

Four steps, and you are the third one.

Connect

Your agent connects to the bridge running on your machine.

Request

It asks to read the tab, use the clipboard, or fill a field.

Decide

You see exactly what it wants to do, and approve or deny.

Log

Every request and decision is written to the activity log.

What it needs

Focused permissions.

A small extension with a short permission list - and nothing beyond it.

Read the active tab

So the agent can see the page you are looking at.

Clipboard read & write

For copy and paste workflows.

Scripting

To type into pages you approve.

No broad “read all websites” permission, no data collection, no third-party servers.

Privacy

Nothing leaves your machine.

The bridge listens on 127.0.0.1 - your machine only. No cloud, no accounts, no telemetry. There is nothing to sign up for.

Getting started

Running in four steps.

InstallAdd AgentBridge from the Chrome Web Store.
Run the bridgeStart the small local process on your machine.
Point your agentConnect it to ws://127.0.0.1:<port>.
ApproveAccept the first prompt and you’re running.

FAQ

Common questions.

Yes. Any agent that can speak WebSocket can use it. There is no vendor lock-in and no proprietary protocol.

Ready to try it?

AgentBridge is free, open source (MIT) - audit the code, run your own bridge, or contribute on GitHub.

Add to Chrome