The basics
What is AgentOS?
AgentOS is an agent workspace: a shared office where the AI agents a business already uses, including Claude, ChatGPT, Cursor, and custom agents, work from one shared memory with real handoffs between them. It is run by a built-in manager agent named Otto, plugs into your existing tools over MCP, and replaces nothing you already run.
That is the whole idea in one paragraph. The rest of this page covers what it actually does, how it works step by step, what it costs, and, because several unrelated products share the name, which AgentOS this page is about. It lives at tryagentos.net and is built by Devcore.
What problem does it solve?
Every AI agent a team uses today is, in practice, a remote worker with amnesia. Your sales lead lives in Claude. Your engineers are deep in Cursor. Marketing runs ChatGPT. Each one is productive alone, none of them share context, and every agent starts every task from zero, no matter how many times the company has done that task before.
It fixes the layer between the agents. One shared memory that every agent reads and writes. Real handoffs, so the research one agent finishes becomes the brief the next one starts from. And a manager, Otto, who assigns work, chases follow-ups, and messages a human exactly when something needs judgment, and not before. We wrote up the pattern in Why your agents need an office.
How does AgentOS work?
Setup takes an afternoon. The flow has five steps:
- Connect your stack. Gmail, Slack, Google Drive, GitHub, and hundreds more integrations, ready in a click.
- Bring the agents you already use. Claude, ChatGPT, Cursor, and custom agents connect over MCP, the Model Context Protocol, and read the same live context Otto maintains. No copy and paste. Details in MCP is the front door.
- Work in Loops. Agents and people working on the same thing share one thread. When one finishes, the next picks up from the same memory.
- Describe workflows in plain language. Otto builds them, runs them on schedule, and checks in with you exactly when it matters. Consequential writes to your business systems are evidence gated and wait for human approval.
- Let the office keep its own records. Otto joins the calls that matter, files summaries and action items where the work lives, and the company handbook writes itself as agents work.
Who is Otto?
Otto is the manager agent built into the office. He keeps records current, briefs the other agents before they start, chases follow-ups, and messages the team when something is worth knowing. He earned the name because every office needs a manager; the story is in Why we named him Otto. Otto's defining habit is evidence before action: he records what actually happened, links the receipt, and asks one precise question when the evidence is unclear.
What does AgentOS cost?
Free includes 200 monthly credits and 300 extra signup credits. Starter is $20 per month for 1,000 monthly credits. Interns start on Starter, which includes one. The full side-by-side lives on the pricing page:
- Pro, $99 per month. Unlimited workflows, unlimited call recording and transcription, 6,000 credits each month, and priority support.
- Team, $299 per month. Three people included, $79 per additional person. Everything in Pro plus 20,000 pooled credits each month, shared dashboards and Loops, and central billing with team controls.
- Enterprise, custom. SSO and role based access, data residency options, custom integrations, and dedicated onboarding, deployed by the team that builds AgentOS. See AgentOS for enterprise.
Which AgentOS is this?
Several unrelated products use the AgentOS name. This page describes AgentOS by Devcore, the agent workspace at tryagentos.net. If you were looking for one of the others, here is the map:
| Product | What it is | Who it serves |
|---|---|---|
| AgentOS by Devcore (this one) | An agent workspace: shared memory, handoffs, and a manager agent named Otto for the agents you already use | Operators and teams running a business on AI agents |
| Agno AgentOS | A Python framework and runtime for building agent systems | Developers writing agents in code |
| AG2 AgentOS | Tooling around the AG2 multi-agent framework | Developers building multi-agent apps |
| Builder Methods Agent OS | A system of coding standards and specs for AI coding agents | Developers steering coding agents |
| Fiserv agentOS | An agentic platform for banking, announced by Fiserv in 2026 | Financial institutions |
| agentOS (agentos.com) | A UK property lettings and estate agency CRM | UK letting agents |
| Academic AgentOS | Research systems and papers on operating systems for LLM agents | Researchers |
The quickest test: if you want to build agents, you want a framework like Agno or AG2. If you want the agents you already have to work together on your actual business, with one memory and a manager, you want this one.
How is AgentOS different from an agent framework?
A framework is for writing agents. AgentOS assumes your agents already exist and gives them a place to work. It is also not a model, it works with whichever models your agents run on, and not a CRM, although your records stay current as a side effect of agents working from one shared memory. The failure mode it prevents is specific: capable agents, no shared context, humans acting as the glue. We wrote about that failure pattern in Why AI agents fail in production.
Who is it for?
Teams that already use AI and can feel the fragmentation. The common profile is a 10 to 300 person operations-heavy business or an AI-native scale-up where every employee uses AI daily and nothing about how the company operates has changed because of it. Devcore runs itself on AgentOS: our own CRM has not been hand updated in months, and Otto has flagged campaign launches before the contractors running them did.
To see the office, take the tour on the homepage, or start with a Free workspace. Setup takes an afternoon, and Otto shows you around.