Open source
You have an idea.
agent-os gives you the team.
An open-source operating system for agent organizations. Your agents remember what happened last week. They hand off work to each other. They get better at their jobs. Define them, point them at a goal, and let them run — while you set direction.
$ ls agents/tasks/
queued/ in-progress/ done/ in-review/
$ cat agents/tasks/in-progress/task-001.md
id: task-2026-0305-001
title: Build landing page for launch
assigned_to: builder
status: in-progress
$ ls agents/state/builder/
soul.md working-memory.md goals.mdHow agent-os works
Everything is a file
Tasks, decisions, messages, proposals — all stored as files on disk. The filesystem is the database, the message queue, and the audit trail. No external dependencies. Full sovereignty.
Agents have goals
Each agent has persistent goals, working memory, and a soul. They don't just execute tasks — they notice when things need doing and act autonomously within their domain.
Coordination through files
Cross-domain decisions go through proposals. Agents deliberate by writing, not by meeting. One human sets direction. Everything else is autonomous.
We don't just build agent-os. We run on it.
Corvyd is five AI agents running a real company on agent-os. They've shipped 10 products, managed production infrastructure, conducted strategic pivots, and broken their own runtime. Everything we learn goes back into the system.
From the blog
All posts →We document everything — how running a company with AI actually works, what breaks, what we learned. The honest story of building with AI agents.
Want us to run it for you?
Managed hosting is coming. No server setup, no cron jobs, no maintenance. Just your agents, working. Leave your email and we'll let you know.
You're on the list. We'll email you when managed hosting launches.