Chat
A model chat interface of the kind you already know, with conversation storage that only the account holding it can open.
The interface
Conversations, a thread view and a composer, with model selection and web search available per conversation. If you have used ChatGPT or Claude, the shape is familiar; the difference is where the conversation is kept and who can open it.
Requests made from the chat product go through the gateway, so the same provider routing applies to what you type here as to anything you send over the API.
Storage and access
Conversations are stored encrypted at rest. The stored form cannot be read with an account's login credentials alone being present: an account's conversations are accessible only to that account, and opening them requires its credentials.
This is the storage guarantee as it applies to the chat product. Terms agreed with the model providers behind each conversation are described on the privacy page.
Accounts, billing and teams
Many users can be associated with a single billing account, so a group of people can be billed as one. Users within a billing account can be divided into teams, and folders can be shared with a team so that its members reach the same conversations.
| Level | What it groups |
|---|---|
| Billing account | One or more users, billed together |
| User | A person with their own credentials and conversations |
| Team | A subset of users within a billing account |
| Shared folder | Conversations visible to a team |
Chat for less than the cost of a Claude subscription
Chat is not sold by the month. You top up a balance and the conversation draws on it, so the only cost is the model requests you actually make plus the 2.5% markup on the provider's list price. There is no seat, no minimum and no fee for a month you barely use.
Credit never expires. A top-up sits on the account until it is spent, and nothing renews in the background waiting to be cancelled.
Claude is around $20 a month, which is $240 a year whether or not you use it. Two things stretch a balance here further: you choose the model per conversation rather than per subscription, and the charge follows each request rather than each month. The most cost-efficient models in the catalogue answer an exchange of a few hundred words each way for a fraction of a cent, so a typical user on those models can expect $10 of credit to cover a year or more.
What a top-up buys depends on how much you write and which models you reach for. Anyone running flagship models hard will spend faster, and a team shares one balance rather than one bill per person. Both the choice and the spending stay in your hands.
Logging
The chat product follows the same policy as the rest of the service: conversations are not logged. Temporary chats leave no trace at all — once the conversation is over there is nothing retained to recover, by you or by us.
Web search
Web search is offered through Bing. When it is used in a conversation, the search results needed to answer the query are retrieved and passed to the model along with your message.
Search is a request to a third party, so it sits outside the provider terms that cover model inference. Treat the search text as you would any query you type into a search engine.
What to read next
The parts of the platform people usually read after this page.