Suits developers comfortable loading an unpacked browser extension and getting an API key or local model server; the README's security docs help if you want to judge the design.

What can you build with peerd?

Automate tasks across tabs where you are already signed in. Run code in a sandboxed JavaScript notebook or Linux VM without touching your machine. Chat with a local model on localhost through a browser-based agent. Study how an agent runtime can be built from browser primitives like workers and WebRTC. Share apps the agent builds with other browsers in a preview build.

NotASithLord /

peerd

Browser extension that runs an AI agent inside Chrome or Firefox using your own key

421 stars
4 stars since 3 Oct
JavaScript
Updated today
Some experience
Browser & computer use

Open on GitHub (opens in a new tab) Website (opens in a new tab)

Most agents either run on your whole computer or in someone else's cloud. peerd takes a third route: the agent lives in a browser extension and relies on the browser's own security boundaries, such as workers, origins, sandboxing and OPFS storage.

The main agent hands work to separate, keyless actors. Each page or compute environment gets one, with tools limited to that environment. Credentials, network rules, confirmations and audit stay with the extension. Code runs in sealed JavaScript notebooks, WASI tools, browser apps, or Linux virtual machines (Chromium only), none of which can reach your host operating system.

You bring your own model: a hosted provider with your key, a local model on localhost, or early WebGPU support. There is no peerd account, and the README says current builds send no product telemetry. Preview builds add browser-to-browser sharing over WebRTC; store packages leave that out.

What it does

Works in your own browser
The agent can read and drive your tabs, web apps and signed-in sessions. The web actor can learn a site once and reuse that client on later tasks.
Sandboxed compute
Scripts, sealed JavaScript Notebooks, compiled WASI tools, browser Apps and Linux WebVMs give the agent local compute without access to the host OS.
Scoped actors
Each page and compute environment gets its own keyless actor. If the browser can't prove the worker isolation boundary, the actor does not run.
Memory and a local vault
Sessions, memory, skills and goals live in the extension. A local vault protects your keys, unlocked by passphrase or passkey where supported.
Choice of model
Supports bring-your-own-key cloud providers and keyless local options. Preview builds also add peer-to-peer sharing over WebRTC.

What you could build

Automate tasks across tabs where you are already signed in.
Run code in a sandboxed JavaScript notebook or Linux VM without touching your machine.
Chat with a local model on localhost through a browser-based agent.
Study how an agent runtime can be built from browser primitives like workers and WebRTC.
Share apps the agent builds with other browsers in a preview build.

Get started

For Chrome, clone the repo, open chrome://extensions, turn on Developer mode and load the extension/ folder unpacked. Then open peerd, create the local vault, add a provider key or choose a local provider, pick a model and start chatting. Firefox needs its own package built with the command below.

bun run package -- --channel=preview --browser=firefox --no-sign

Commands are copied from the project's README. Read it on GitHub for the full setup.