All work
AI & automationOpen source

AI Browser Agent

A Chrome side-panel agent with a FastAPI planning backend, JWT auth, a secure credential vault and a step-by-step browser execution loop.

Product
Chrome extension + AI backend
Built with
FastAPI, LangGraph, PostgreSQL
Source on GitHub
AI Browser Agent reading a LinkedIn profile in the side panel
Step by step

plan, act, observe, continue

0

credentials stored in the extension

1 click

to stop the agent at any time

Overview

The problem we set out to solve.

A browser automation product split in two: the backend plans actions step by step, and the extension carries them out in the user's own browser. That split makes the agent feel deliberate instead of random.

  • Browser agents often act unpredictably and are hard to trust.
  • Users need to see what the agent is doing, and stop it instantly.
  • Saved credentials must never live in the extension itself.
The agent's answer summarising the profile
Each step with its tool output: navigate, wait, read page
Execution timeline with step status
What we built

Every part, built to work in the real world.

Step-by-step loop

A LangGraph agent loop that can resume from the last observed tool output.

Reads, not guesses

Reads the page as structured data, each element's role and name, instead of guessing from pixels.

Local execution

Acts through the Chrome debugger, tabs and scripting APIs in the user's real session.

Credential vault

Saved logins for login-assist flows, kept server-side.

Transparent timeline

Every action shown live with its status, and a stop button while the agent works.

Validated decisions

Structured tool decisions are checked by the backend before anything touches the page.

How it's built

The FastAPI backend plans and continues from tool outputs; the extension executes. Server-side guardrails constrain which browser tools the agent can use, so it always stays inside a defined envelope.

  • FastAPIPlanning backend, auth and run state
  • LangGraph & LangChainAgent loop and guardrails
  • PostgreSQLUsers, sessions and run history
  • Chrome Extension APIsDebugger, tabs and scripting
  • JWTAuth across extension and API
The outcome

A browser automation assistant with secure auth, saved run state and an execution view clear enough to review and trust.

The five pipeline stages, from input video to structured JSON
Next project

Video Verification Pipeline

AI document pipeline

Have an idea? Let's build it.

Tell us what you want to build. You'll get a clear plan, a timeline and a fixed price within a week.