Launching today
Pixel Office
Run a team of AI agents from a pixel-art office
59 followers
Run a team of AI agents from a pixel-art office
59 followers
Most agent tools are a chat box. Pixel Office is a place: each employee is a real AI agent (Claude, Codex, Gemini or OpenRouter) with its own role, memory, skills and permission level. See at a glance who is working, who needs your approval and who is done. Hire in a character creator, hand out tasks on a board, let agents message each other, and decorate the office with coins from finished work. Open source, one command, runs on your machine with your own keys.
Free
Launch Team
Render Sandboxes: a computer for every agentRun untrusted code safely. Idle sandboxes cost nothing
Promoted
Hi Product Hunt! I'm the maker of Pixel Office.
I was running several AI agents at once and kept losing track of them: which one is working, which one is stuck waiting for my permission, which one finished an hour ago. Terminal tabs and chat boxes all look the same. So I gave them an office.
Each employee is a real agent (Claude, Codex, Gemini or OpenRouter) with its own job description, memory file, skills, model and permission level, working in the folder you give it. You hire them in a character creator, hand out work on a board, and watch them walk to their desks. They can message each other to hand off work, hold meetings with a written summary, and the coins from finished tasks go to furniture in the Market.
It is open source (MIT) and runs on your own machine with your own keys: npx github:basriayaz/pixel-office
I'd love feedback on two things: is the first run clear enough, and what would you want an office of agents to do on its own while you are away?
how do u tell from the office view if an agent is stuck or just taking a while
@igorblink Good question! The board flags any task whose owner is waiting for your permission, hit a session error, stopped working or left the office, and the 📋 button shows how many need you. If someone is just sitting at their desk, it's almost always a pending permission card in their chat, and turning on 🔔 notifies you when that happens. An agent that's still mid-task stays "working" on the board, so slow and stuck look different at a glance.
This is the exact problem I keep hitting — I run several Claude Code sessions at once and lose track of which one is doing what, and which one is stuck waiting on me. Giving each agent a desk, a job title and its own permission level makes that state visible instead of a pile of terminals. One question: does each employee's memory persist between sessions, or does it start fresh every time it sits down at its desk?
@jcbuilds Thanks JC, that's the exact pain this came from. It persists: each employee has a plain Markdown memory file that's loaded into their system prompt at the start of every session, and they append lessons after each task. Resetting a chat keeps the memory, and you can open and edit the file yourself from the profile page.
Such a good idea to keep track of agents, giving them a personality based on what work they do too really helps!
@katie_harris7 Thanks Katie! That was exactly the idea: once each agent has a face, a desk and a job title, you stop asking "which terminal was that again?" Glad the personalities land 🙂