Your agents
Chatting with your AI agents, the templates they start from, the front page prompt box, and what a conversation costs.
An agent is an AI assistant with a name, a persona, a model and a set of tools. Your org can have several, each for a job.
Chat
/agent — pick an agent and talk. The conversation is kept as a session, so you can return to it. When an agent uses a tool (creates a page, checks the calendar, searches the web) it says so in the reply.
Your Butler
Every account has one agent already: the Butler, the assistant behind the bow-tie icon in the top bar of every page. It reads the help pages, knows which page you are on, and can look after your tasks, calendar and mail. See Your Butler.
Starting from a template
/agents → Templates gives you a ready-made agent in one click:
- Personal Assistant — tasks, calendar, email, search.
- Builder — creates projects, builds pages, adds products, invites users; assembles and runs a business end to end.
- Project Architect — chat through an idea and it sets up the right project.
- Project Manager, Support Concierge, Calendar Concierge, Social Media Helper, Research Assistant, Email Assistant, Coder, and Orchestrator (plans and delegates to specialists).
A template is imported once; after that it is your agent to rename and reshape.
The front page prompt box
Sites with a prompt box on the front page hand what you type to your Builder as its first message — it is created for you the first time. Type the business you want and the conversation starts with it already working.
What a conversation costs
Replies are metered in service credits by the model used and the tokens exchanged, unless you have connected your own key for that provider — then they are free. A run stops cleanly when your balance would not cover the next call. See Service credits.
Where else agents live
- Telegram — talk to the same agent from your phone. See Channels.
- Projects — assign a task to an agent and it works it in the background. See Goals and the build engine.
- Flows — a step can hand work to an agent and continue with its answer. See Step types.