03Browser agent Chrome extension
Do It For Me (DIFM)
A Chrome extension that handles repetitive web chores: it reads rough notes, opens the right sites, fills forms, tracks prices and preps checkouts inside your browser.
4
packages in a pnpm monorepo
MV3
Chrome Manifest V3 extension
1-click
manual confirmation on payments
Overview
Instead of hardcoded scripts, DIFM reads the browser’s accessibility tree, hands it to an LLM planner, and executes the next action the way a person would. It prepares everything on screen but always asks for a one-click confirmation before sensitive steps like final payments.
(01)Features
What it does
Smart task parsing
Drop a messy note or a screenshot of a bill. The agent extracts the biller, consumer ID, amount and due date and schedules the task.
Autonomous execution
At the right time it opens the site, finds the form fields, types your details and clicks through to the final checkout screen.
Price tracking and carting
Tell it to buy something below a target price. It watches the page in the background and adds the item to the cart the moment it qualifies.
Identity vault
Personal, work and household profiles, so the right name, email and address are used for each task.
(02)Architecture
Inside the monorepo
- 1
apps/extension
The Manifest V3 extension built with WXT and Preact: side panel UI, content script and a DOM executor for click, type and scroll.
- 2
packages/a11y-tree
Turns the DOM into a lightweight semantic accessibility tree with specialised extractors for e-commerce prices and cart buttons.
- 3
apps/server
The planner service. It takes the goal and the page tree and returns the next action, with support for several LLM backends.
- 4
packages/shared
Shared TypeScript types, Zod schemas and protocols used across the extension, server and libraries.
(03)Tech stack
Built with
Extension
- TypeScript
- WXT
- Preact
- Manifest V3
Planner
- LLM planner
- OpenAI
- Qwen
- Zod
Tooling
- pnpm workspaces
- Vitest
Next project