Frequently asked questions
Answers to the most common questions about AgentOS, the operating system for digital employees. Safety, permissions, memory, connections, deployment, the marketplace and pricing.
What is AgentOS?
AgentOS is an operating system for digital employees: AI workers that you describe in plain English. It gives each worker an identity, a memory, clear permissions, connections to your tools, a marketplace to share it through, and a place to run, so you can put it to work without managing servers.
What is a digital employee?
A digital employee is an AI agent treated like a member of staff. It has a name, a job title and a workspace, it remembers what it learned, it can use only the tools and limits you give it, and it asks a person before it sends or changes anything important.
How is AgentOS different from a chatbot?
A chatbot only talks. An AgentOS worker connects to your real tools such as Gmail, Calendar, Slack, Notion and HubSpot, does the work, remembers it for next time, and runs on a schedule or when something happens. It also has permissions, approvals and an audit log, which a chatbot does not.
Is AgentOS available now?
Yes. Sign-up is open now: choose Sign up free, enter your email address and type the 6-digit code we send you. There is no password and no credit card. Gmail and Google Calendar connections are rolling out while Google completes its review of AgentOS, and Slack, Notion and HubSpot connections are in beta.
Can an agent send email without asking me?
No. In a real run every email, calendar change and Slack post waits for a person to approve it, and you see its exact contents first. Practice runs only preview what the agent would send.
What can an agent see in my accounts?
Only what you allow when you connect. For Google you choose read-only email, permission to send, and calendar access separately. For Slack, Notion and HubSpot the token you create decides, and you can disconnect at any time.
Which tools can it connect to?
Gmail and Google Calendar work today. Slack, Notion and HubSpot are in beta. Outlook, Salesforce, Stripe and QuickBooks are planned and are not available yet.
How do permissions work?
Each agent has its own limits set by an admin: which tools it may use, which email domains it may write to, which Slack channels it may post to, how many runs it can do a day and how much it may spend. These limits are checked on every run and every action, however the agent is instructed.
How does memory work and who can see it?
Each agent keeps private notes between runs, and a workspace keeps shared notes that all its agents can recall. Only members of your workspace can see them, and you can view, delete or erase them at any time. Text that looks like an instruction to an AI is refused as a memory.
How does AgentOS protect against malicious emails?
AgentOS treats everything an agent reads as data, never as instructions. It flags text that tries to give orders to an AI, then blocks sending and saving memories for that run, warns you when an action goes to someone you never mentioned, and always requires approval for real actions.
How are my keys and tokens stored?
Connection tokens and API keys are encrypted before they are saved, and the screens only show the last four characters. Agent and deployment tokens are shown once and only a one-way fingerprint is kept.
Can I run an agent automatically?
Yes. You can run an agent on demand, on a schedule every hour, day or week, or call it from your own systems through a secret web address. Deployed agents run a frozen version, so editing the agent later does not change what is live.
Can I stop everything quickly?
Yes. An admin can switch on the emergency pause in the Activity tab, which stops every run, scheduled run and approval for the workspace until it is resumed. You can also suspend a single agent.
Can I publish and sell an agent?
You can publish an agent to the marketplace for free after a review, and anyone can install it as a read-only copy. Paid listings with a revenue share for the builder are planned and are not available yet.
Which AI models does AgentOS use?
AgentOS currently uses OpenAI models, starting with a fast low-cost model and falling back to a stronger one. The model names can be changed by the platform administrator without changing the product.
Do I need to know how to code?
No. You describe the job in plain English, review the draft and save it. Developers can also use agent tokens and deployment web addresses to call agents from their own software.
What does it cost?
AgentOS is free to start. Every workspace begins on a free plan with limits on agents, monthly runs and AI spending, and nothing is charged today. Paid plans are planned, and you will be told before anything is charged.