Day 54 of 2026 building!

I built a virtual office where AI agents think, collaborate, and hire their own team.

AgentOffice is an open-source TypeScript project where LLM-powered pixel characters have their own workspace. They walk to desks, talk to colleagues, assign tasks, execute code, search the web and when they decide the team needs more people, they hire new agents on their own.

The entire thing runs locally using Ollama. No cloud APIs. No vendor lock-in.

How it works:

Each agent runs a perceive → think → act → remember loop every ~15 seconds. The LLM returns structured JSON decisions - move somewhere, talk to a colleague, run a code snippet, or hire a new team member.

Memories persist in SQLite with Ollama-generated embeddings, so agents remember conversations and decisions across server restarts. When an agent hires someone new, the recruit spawns at the office door, walks to an available desk, and starts their own AI-powered work cycle.

Why I built it:

I wanted to explore what happens when you give AI agents a physical space, persistent memory, and the ability to grow their own team. The results are surprisingly organic - agents develop working relationships, remember past conversations, and make autonomous decisions about when to hire.

The project is designed for forking. Swap the office for a classroom, hospital, game world, or DevOps war room. Change agent personalities. Add custom tools. Deploy with a single `docker compose up`.

Some ideas:

🎓 AI Classroom - teacher and student agents

🏥 Virtual Hospital - doctor, nurse, patient workflow

🎮 NPC Engine - agent brains for your game

📊 Startup Simulator - watch strategy emerge

🧪 Social Experiment - vary personality traits, observe behavior

Github link in the first comment below!