Skip to main content
You choose a model in MITHUNAI Chat from the model menu at the top of a thread. Every model in that menu comes through your organisation’s Model Gateway: Chat holds no provider key of its own, and every message, every tool step and every generated thread title is a Gateway request, counted against your organisation’s Gateway usage and budget.

​
Pick a model

  1. Open the model menu at the top of the thread. The models are listed under Models.
  2. Choose one. The next message in this thread uses it.
You can change model part-way through a thread; earlier replies keep the model that wrote them. A new thread opens on your deployment’s default model. Your administrator decides how long the menu is. A deployment with no named model opens on auto and lists everything the Gateway offers. A deployment that names one or more models lists exactly those, and opens on the first.

​
What auto means

auto is the Gateway’s own router: it picks a model and a provider for each request and, if that provider fails, retries on another. It works from the list each provider key really has: the Gateway reads that list every six hours, or when an administrator presses Fetch models, and remembers for a day anything the provider refused. A model your account is not entitled to can still be tried first, so a message can occasionally fail. If a message fails with “not found” or a refusal from the provider, send it again or choose a named model you know works, and ask your administrator to name it for your deployment. See Routing and failover.

​
Why a model is missing

Unless your deployment names its models, the menu lists what the Gateway offers your organisation, read from the Gateway when Chat loads it. A model is missing when:
  • Your deployment names its models and this is not one of them. Ask your administrator.
  • No provider for it is set up in the Gateway. Providers and their keys are added in the Gateway, not in Chat.
  • The provider credential behind it is scoped elsewhere. A credential in the Gateway can be limited to a named list of models, and is never used for one outside it.
  • It was added after you opened Chat. Reload the page.
You cannot paste your own provider key into Chat. That is deliberate: one place holds the keys, and every token is attributed to your organisation.

​
How usage is counted

Separately, each answer the Knowledge tool’s ask_assistant gives counts against your organisation’s monthly answers, the same allowance as the console, the widget and the API. The Gateway’s usage and cost figures cover Chat’s traffic only: the console, the widget and the API reach models through the platform directly. They are not the place to read your organisation’s total spend.

​
Where your messages go

The text of each turn — your message, the thread so far, any attached file’s text, and what the Knowledge tool returned — is sent through the Gateway to the provider of the model you chose, under that provider’s terms. MITHUNAI does not train models on your conversations. See Security.
Last modified on November 7, 2026