Loading the documentation…
Every chat answers with one model. You choose it in the message box, set how much thinking effort it uses, and switch only where the chat allows it. The models a chat can offer come from the Mind's AI Config and your company's AI configuration; see Choose models and AI behaviour. To send messages and read answers, start with Ask a Mind.
On a phone, open + under the message box and tap the Model row; the same picker opens as a step inside the drawer. The message box and its controls are described in The message box and its settings.
The bottom of the picker holds the Effort control with the current value. Point to it to open the effort choices on the right; on a phone, the Effort section sits under the Model row in the drawer.
When you switch models, your choice stays only if the new model offers the same level; otherwise the chat returns to the new model's default.
A chat fixes its model at the first message. This rule is for chats with a Mind in DataMind OS; in DataMind Oda you can switch the model of a chat at any time (Choose the model for a chat). A chat started on Kartlos Premium or Kartlos Standard stays on exactly that model. A chat started on another model can move between other models, but it cannot switch to a Kartlos model. To answer with a different model, start a new chat.
In the picker, a model the chat cannot use is greyed out and unselectable. Point to it to read why; on a phone, the reason shows under the row.
If a send carries a model the chat cannot use, the chat answers with its own model and shows the notice Model locked for this conversation. The picker also returns to the model the chat uses.
Sending is blocked while the chat's model cannot be used. The message box shows a badge, and the picker stays open so you can switch and recover. Point to the badge to read the full explanation; on a phone, tap it.
A message is always sent with one model. If that model cannot be used, the message is refused so you can pick another; another model is never swapped in silently. Your text comes back to the message box, so choose a model from the list and send again. The chat explains the refusal with one of these lines:
| What the chat says | What it means |
|---|---|
| "No model was selected. Pick a model and send again." | The message carried no model. |
| "This model is not on this mind's list. Pick one from the model menu, or ask a mind manager to enable it." | The model is not enabled for this Mind. |
| "An administrator turned this model off for your organisation. Pick another model, or ask an admin to allow it again." | Your company no longer allows the model; the picker drops it as well. |
| "This model is no longer in the catalog. Pick a different model and send again." | The model has left the catalogue. |
Enabling models for a Mind is described in Choose models and AI behaviour.
Sometimes an answer never starts because the AI service behind the chat could not take the request. The chat shows a short message in place of an answer:
| What the chat says | What it means and what to do |
|---|---|
| "Insufficient credits" | The account has no credits left for the request. Ask your administrator to check the account, then send again. |
| "Model requires moderation, contact support" | The request must pass moderation before a model can take it. Contact support. |
| "No provider could serve this request with the selected model", asking you to switch to a different model | The chosen model could not be served for this request. Pick another model from the list and send again. |
| The chat reports an invalid API key. | A problem on the service side, not something in your message. Try again; if it keeps happening, tell your administrator. |
| The chat reports that the AI service is temporarily unavailable. | The service is paused for a moment. Wait a little, then send again. |
| "Model is down, please choose a different one or try again later" | The chosen model is not answering right now. Pick another model, or try again later. |
| "There is no available model provider that meets your routing requirements" | No available provider matches the routing requirements for the request. Pick another model, or try again later. |
More messages and their fixes are listed in Minds troubleshooting.
A reply can stop part-way when the model reaches its output limit for a single answer. Everything it wrote stays in the chat, and this message follows:
"The response hit the output limit and stopped early. What was written so far is kept - press Continue to pick up where it left off."
Press Continue to let the model carry on from where it stopped; the retry and continue controls are described in Chat states and recovery.
The AI Models Comparison window lists every model in the catalogue side by side. Each card shows the model's tier, Standard or Premium, a New badge where it applies, its Capabilities (Tools, Reasoning, Structured Output, Images, Files) and its Details: Context, Max Output and Reasoning Efforts. The window counts the models it shows, and reads "No models available to compare." when there are none.
To compare models on your own questions, use Test a Mind with benchmarks: each experiment runs the same questions against variants that differ in model, thinking effort and response mode.
Click the model's name in the message box and pick from the list. On a phone, open + and tap the Model row. The list holds the models the Mind offers; see Choose models and AI behaviour.
No model is set for the chat yet. Pick one from the list; the first message you send fixes the chat's model. If you send without one, the message is refused with "No model was selected. Pick a model and send again." and your text comes back to the message box.
The chat cannot use that model. Point to the row, or read under it on a phone, for the reason: the chat is locked to another model, it started on a non-Kartlos model and cannot switch to Kartlos, or the model cannot read the images or PDFs the chat already holds.
A send carried a model the chat cannot use, so the platform answered with the chat's own model and told you. The picker also returns to that model; your next messages use it. To answer with a different model, start a new chat.
Model retired means the model is gone from the catalogue; Model unavailable means it is not activated for this Mind; Model incompatible means it cannot read images or PDFs in the chat. Sending is blocked while a badge shows; the badge text names the fix, and the picker stays open so you can switch.
Your effort choice is kept only when the new model offers the same level. Otherwise the chat returns to the new model's default.
The model has no thinking options, so there is nothing to set. Switch to a model that supports thinking effort.
The list holds only the models your company offers and the Mind allows, so it can be shorter than the full catalogue. A Mind manager adds models on the Mind's AI Config tab; see Choose models and AI behaviour.
Every model in the catalogue, side by side: its tier (Standard or Premium), a New badge where it applies, its Capabilities (Tools, Reasoning, Structured Output, Images, Files) and its Details (Context, Max Output, Reasoning Efforts). To compare models on your own questions, see Test a Mind with benchmarks.
The Mind offers only the models enabled for it, and this one is not among them. Pick one from the model menu, or ask a mind manager to enable the model. See Choose models and AI behaviour.
The AI service behind the chat could not take the request, so no answer started. "Insufficient credits" means the account has no credits for the request: ask your administrator to check it. "Model is down" means the chosen model is not answering right now: pick another model, or try again later. When the service cannot take the request lists every such message.
The model reached its output limit for a single answer, so the reply stopped part-way. What it already wrote stays in the chat; press Continue to let it carry on from where it stopped.