Launching today
OpenVorld
Give your AI agents a home you can see
78 followers
Give your AI agents a home you can see
78 followers
OpenVorld is an open-source harness for AI agents, with a visible world on top. Every project is a pixel-art village, and every agent has a name, a role and a brain: Claude Code, Codex or any OpenAI-compatible API. You watch the crew work, and they ask before they use a folder, a command or a website. Free, open source and local. No account needed.










Free
Launch Team

Memanto CloudYour agents focus. Memanto remembers.
Promoted
Hi Product Hunt, I'm Amir. Thanks for stopping by!
OpenVorld is an open-source harness for AI agents. It runs agents like Claude Code or Codex as a crew, and shows the work as a small pixel-art village.
The problem: AI agents can do real work now. But most of them live in a terminal, and that makes them hard to trust. Text scrolls past. You can't tell who did what. You wonder if they will touch the wrong file or waste your tokens.
Why we need something different: if you can't see your agents, you can't manage them. And the more agents you use, the worse it gets. I built OpenVorld to make the work visible, so it feels like running a small team and not like reading logs.
How it works:
Every project is a "vorld", a small pixel-art village.
Each agent has a name, a role, a personality and a house. You can see their activity above their heads.
You are the founder. Agents only act when you give them a goal, and they ask you before they use a folder, a command or a site.
Bring your own brain: Claude Code, Codex, or any OpenAI-compatible API like OpenRouter or Ollama. Each agent can use a different one.
Finished work gives XP, levels and deeds. Work should be a little fun too.
Your files stay yours. Everything is saved on your computer. There is no account and no hosted backend. It's free and open source.
Links:
Try the web demo, no download: openvorld.com/play
Source code: github.com/DigitalDreams-Lab/OpenVorld
Website and downloads: openvorld.com
It's still early, and I'd love your honest feedback. Which job would you give your crew first?
first job id give it is dependency bumps and fixing whatever tests break after, boring stuff i always put off. can two agents work in the same folder at once or does one wait for the other?
@bacelj Hey David, thanks for your comment.
I guess so. One of my initial goals is to make party creation easy for single-use purposes or long-term, where you can set a goal and let multiple agents (from different harnesses) work together to deliver your goal.
Every agent has its own private workbench(workspace), and you can define multiple workbenches, including a common one between all agents to work on a shared area. Plus, in parties, there is a shared artifact folder where every agent can create/edit a file. They can put resources there so every agent can easily reach and analyse them.
From fixing a bug to ideating for a new feature, you can create collaboration between them.
I'm working day to day to improve this experience. Since it is open source, you can check it and even create a PR to add any missing features.
You can mention any agents or mention @all to let everyone respond to you.
Asking before a folder, a command or a website is the part I'd use first. Where it wore me down in my own setup was the second week: the same approvals again in every new session, until I either clicked yes without reading or opened everything up. Does a villager keep what it was allowed last time, and can you see in the village which permissions it is still carrying?
@heinrich_neb Hey Heinrich, thanks for your comment.
Good question, Heinrich. I designed a new writ system that you can use to manage allowed/revoked permissions for every AI agent, so it stays per agent rather than per session.