HIVEMIND - Swarm Intelligence for Your Browser

A swarm of browser agents that automates tasks on your behalf, just one iMessage away.

Comment

Inspiration

Most browser agents are slow because they process screen images one by one. I wanted to build something faster and more powerful—a swarm intelligence system that works inside your real browser. Hive Mind was born from the idea that one agent isn't enough for complex web tasks. You need a coordinated swarm.

What it does

Hive Mind is an AI-powered browser orchestration system. You can give it tasks by typing, speaking, or even texting via iMessage. It deploys a swarm of agents that work in parallel across your real Chrome tabs. Whether you're comparing flight prices across multiple platforms or applying for jobs from your phone, Hive Mind handles the heavy lifting. It attaches to your actual browser session, so it's fast, token-efficient, and bypasses bot detection.

How we built it

We used the MiniMax M2.7 model as our 'Queen' orchestrator. It handles intent classification and routes tasks to specialized worker agents. These workers control the browser via CDP (Chrome DevTools Protocol), allowing them to perform actions simultaneously. We integrated Supermemory RAG so the swarm can make decisions based on your personal history. For the mobile experience, we built an iMessage bridge that lets you trigger complex browser workflows with a simple text.

We leveraged TRAE AI extensively for rapid prototyping and full-stack integration. Specifically, TRAE was used to generate and iterate on our "Hexagonal Hive Dashboard" UI in React/JSX, allowing us to seamlessly integrate our playful "Swarm" visual design with the complex underlying WebSocket and Zustand state management without breaking the core logic.

Key Features

  • Parallel Execution: Compare products on Amazon and Best Buy at the exact same time.
  • iMessage Control: Apply for jobs or fetch email summaries by texting the swarm.
  • Human-in-the-Loop: The system warns you about sensitive actions and asks for approval before sending emails.
  • Master Dashboard: You act as the admin, viewing the reasoning of each agent and taking control whenever you want.
  • Scheduling: Set tasks like 'fetch unread emails daily at 9 am' to get automated summaries.

Built With

  • agent
  • cdp
  • llm
  • minimax
  • nextjs
  • python
  • trae
Share this project:

Updates