Which AI models the agent uses
Two different things are often confused: the model the agent thinks with, and the models it uses as tools. They are chosen — and paid for — differently.
What runs the agent's reasoning?
An AI subscription that belongs to you. Two providers are supported today: Anthropic Claude and ChatGPT (a Plus or Pro subscription). The subscription is issued in your name and stays yours — we do not resell it and it is not bundled into the plan; on the Assisted plan we connect it for you.
Can one agent have several subscriptions?
Yes, and it is the most useful setting on the agent's page. Subscriptions are an ordered list: the top one is in use, the next one is the standby. Turn on switch to the standby when the limit is hit and the agent keeps working instead of waiting for a reset.
You can connect as many Claude subscriptions as you need. ChatGPT is limited to one per agent. Adding one is done in the dashboard on the agent's page — you get a one-time link, sign in there, and the credentials never pass through a chat. See connecting your subscription.
Can I make one chat cheaper, or one chat smarter?
Yes, and the model is set per chat, not per agent. If a chat is mostly short confirmations and lookups, a lighter model handles it at a fraction of the cost. If a chat is architecture, code or long multi-step work, the strongest model earns its keep there.
You do not need the dashboard for this — tell the agent in that chat: "work on a cheaper model here" or "this chat is heavy, use the strongest one". It switches itself, and its memory of the chat is untouched. A good agent will suggest the switch itself once it sees that a chat is consistently simple.
What about images, video, voice and other tasks?
For generating pictures, video, music, speech and 3D, and for specialised jobs like transcription, the agent does not use its reasoning model at all — it calls a hub of purpose-built models through a single gateway and picks the right one per task. That allowance is part of your plan; the VIP plan carries a five times larger one.
What about Gemini, or another provider?
Not today — the two brains above are what the platform runs on right now. If another provider matters for your work, write to support@emagi.ai and say which one and why; that is how we decide what to build next.
Do I buy tokens from you?
No. The reasoning runs on your own subscription, in your name. What you pay us for is the agent itself: its isolated server, its memory, its skills and the platform around it. See what you pay for.