Vol. I · No. 579Rochester, New York, Thursday, September 24, 2026Multi-agent · Page B17
All workSection B · Page B17

Multi-agentPixelHQ

Hire AI Employees, Then Walk Over to Their Desks

A pixel-art office where every AI agent is an employee with its own desk, skills, scheduled jobs and Docker container.

Graphic5 steps
  1. OfficePhaser scene you walk around with WASD
  2. HireCreate an agent employee from the sidebar
  3. ContainerDocker starts an OpenClaw runtime for it
  4. WorkChat over WebSocket, cron jobs on a schedule
  5. KnowShared ChromaDB knowledge base
What happens inside PixelHQ, in order.Graphic: The Sahil Bachu

Managing a bunch of AI agents usually means a wall of terminals. I wanted it to feel like walking around an office. PixelHQ is a pixel-art office where each agent is an employee. You walk up to one, click to chat, hand it skills and scheduled jobs, and the whole team shares a knowledge base.

It's a fork of SkyOffice, Kevin Shen's open-source multiplayer office built on Phaser, and the pixel art is by LimeZu. My first commit ripped out what I didn't need: multiplayer rooms, WebRTC video, screen share, the whiteboard and the chat stores. It touched 82 files. In their place went React sidebars for agents, an agent-creation dialog and a Redux store for agent state.

One container per agent made isolation easy and orchestration hard.

The backend took most of my time. It's Express and TypeScript over SQLite, and every employee gets its own OpenClaw container that Docker starts through dockerode, with a free port and tracked state. A service keeps a WebSocket open to each container and reconnects when one drops. node-cron runs scheduled jobs and streams their history over Socket.IO. Skills install per employee, with their credentials encrypted.

The shared knowledge base is ChromaDB with Hugging Face embeddings, and a research service can search Wikipedia and scrape pages into it. Agents can run on OpenAI, Anthropic, Gemini or Groq behind one provider layer. Login is Google OAuth with JWTs.

I stopped working on it in February 2026. Most of the backend never made it into a commit, so the public repo only shows the first refactor. If I pick it back up, the container setup is the first thing I'd rethink: one container per agent made isolation easy and orchestration hard.

More from The Work

See all 25

Agentic AI

Your Coding Agent Fills the Gaps in Your Resume

A local-first orchestrator that reads a student's resume, proposes projects to fill the gaps and has their own Claude Code or Codex build and deploy them.

ResumeGod reads your resume, suggests projects that would fill its gaps, then drives your own Claude Code or Codex to build and deploy them. It finishes with a personal site and a fresh LaTeX resume. Nothing is hosted centrally.

Continued on Page B15

Generative UI

A Canvas of Widgets You Order in Plain English

An infinite canvas where an AI agent turns a plain-English request into a live widget wired to your own data.

Ask for something like a card that summarizes your unread emails, and an agent researches, designs and places a live widget on an infinite, zoomable grid. Connectors for Gmail, Calendar, Slack, GitHub, Linear and Notion feed it real data.

Continued on Page B11

Browser Agents

The Job Hunt Agent That Refuses to Hit Submit

JobOps finds jobs, tailors a resume for each one and pre-fills the application, then stops and waits for me.

A job-search command center that runs on my laptop. It pulls postings from five sources, scores them with headless Claude, tailors a LaTeX resume per job and pre-fills applications in Chrome. It never submits. I review and send every one.

Continued on Page B9