Drover by Kirigaya Shirogane

Drover turns the browser you are already using into something a local program can drive.

Every other browser-automation tool starts its own browser: a fresh profile, no cookies, no logins, launched behind a debug flag. Drover works the other way round — it runs inside the browser you already have open, so whatever drives it sees the sites you are already signed into.

How it works

Drover opens a WebSocket to a program running on your own computer, at an address you type in yourself, which must be a loopback one such as ws://127.0.0.1:8790. That program sends requests; Drover carries them out in the active tab and sends back results and events.

It is a driver, not an agent. Drover contains no model, no planner and no task loop, and never acts on its own. Every action comes from the program you connected it to.

What it can do

• Read a page's interactive elements — roles, labels, bounds, validation state
• Click, fill, press, select, check, uncheck
• Navigate, wait for load, screenshot the visible tab
• List, activate, open and close tabs
• Push events: tab opened and closed, navigation, downloads started and finished

What you need

Drover does nothing on its own. You need a companion program on the same machine that speaks its protocol. The protocol is fully documented and anyone can write one — see the repository, which also ships a dependency-free example server you can run in one command. A working open-source client is TENKA, a local voice assistant.

Privacy and safety

• Nothing is ever sent off your machine. The only network destination is the loopback address you enter.
• Connections from anything that is not an extension are refused before the token is read.
• One connection at a time. A second client is refused rather than swapped in.
• A bearer token that you paste in yourself.
• It never intercepts, blocks or modifies network traffic.

Source, protocol documentation and issue tracker: https://github.com/Om-Khode/Drover