Models

You choose the model in the chat's message field. You can change it at any time, even in the middle of a chat. Every reply shows which model wrote it.

1. 1

   Open the model picker

   Click the name of the current model in the message field, or Choose model. On a phone the button shows only a logo, without the name.
2. 2

   Compare models

   Hover over an entry or move to it with the keyboard. An info card about that model appears beside the list.
3. 3

   Pick the model

   Click the model or Auto. The current choice is highlighted in the list, and your next message goes to this model.

Auto sits at the top of the model picker. Behind it is the default model an admin has set for new chats. If nobody has set one, Custodos uses its own default model. For that it picks a model that can use tools, as long as your workspace allows one.

The line under Auto tells you which model it currently stands for. If you have no special requirements, stay on Auto. Another model can be worth it for very long documents, for demanding analyses or when you want to compare two replies. To compare, use Regenerate with … on the latest reply, see Chat basics.

The model picker sorts the models into groups. The group tells you who developed the model and where it runs.

The info card beside the list describes the model your mouse or keyboard is on. On narrow screens it is hidden.

• Provider and name, plus a short description of what the model is good for.
• Credit usage in three levels: Low, Medium or High, with the approximate credits per reply underneath. The figure assumes one question and one reply of average length. How credits work is explained under Credits and throttling.
• Context window in tokens: how much text the model can handle at once in a chat, attachments included.
• Hosting location: the country or region and the infrastructure provider the model runs on.
• Notes when a model thinks before answering or cannot use tools (see below).

Some models cannot use tools. They answer questions and read attached files, but they do not search the web or the Company Brain, and they do not create or edit files. The info card says so with the sentence No web or brain search, and it creates no files.

When a Brain is attached to the chat, these models are greyed out in the picker and cannot be chosen. The info card still explains why.

Some models think before they answer. Until the first text appears, the chat shows Thinking… for them. The thinking is billed like output, so these replies use more credits than their length suggests. The info card points this out for these models.

The table lists the chat models in the Custodos catalogue as of 28 September 2026. Which of them you see depends on your workspace's processing region and on what your admin has allowed. The model picker is always the authority.

Admins use the Models area to set the region the workspace is processed in, which models are allowed and which model new chats get. Members choose freely among the allowed models in the chat. How this works is described under Managing models.

If no model is allowed yet, the picker shows No model enabled yet. Ask an admin to enable one.