Use AI Config to choose the models a Mind offers and the one it uses by default, set how long and how much memory it may use, and switch its features on or off.
The tab has three sections: AI Models, AI Behavior & Features and Similar search. The same settings are also reachable from the AI Config tab of the Edit Mind window, which shows a smaller set of controls: Max Memory Usage and Similar search appear only on the Mind's own tab. Other Mind fields, such as its name and colour, live on the Settings tab; see Mind settings.
Your company's AI configuration decides which models can be offered, so the list may be shorter than the full catalogue; see AI configuration. Kartlos Premium is always available and is the default for a new Mind.
A Mind must keep at least one model. With none switched on, the save is refused with At least one model must be available.
A model that your administrator has switched off, or that is no longer in the live catalogue, cannot be saved. The refusal names the reason: "Model(s) disabled by your administrator: [names]. Ask an admin to allow them, or pick from the available model list." or "Unknown model(s): [names]. Pick from the live model catalog."
Users pick a model per chat, and the Mind's default answers when they do not choose one. See Models in a chat.
| Setting | What it does | Default |
|---|---|---|
| Max Execution Time | The longest time the Mind may take to execute, from 5 to 200 seconds, in steps of 5. | 100 seconds |
| Max Memory Usage | The most memory the Mind may use, from 1 GB up to the platform maximum. | 4 GB |
The Max Memory Usage slider stops at the platform maximum. Hover over the info icon next to a label to read its help text.
Web Search, Context Management and Swarm each have a switch and a Default button that makes the feature start switched on in every new chat. On Swarm, the default shows as an enabled by default pill next to Parallel sub-agents; turning the feature off clears it.
Similar search applies only to vectorized training cards, the cards a Mind finds by similarity search instead of reading them every time. See Train a Mind with training cards.
| Setting | What it does | Default |
|---|---|---|
| Similar results limit | The most passages retrieved in one search, from 1 to 50. | 5 |
| Minimum similarity score | Passages that score below this are left out, from 0 to 1. | 0.4 |
Only Warehouse Builder and Lake Builder Minds show Default change acceptance. It sets the mode a new chat starts in.
Unrestricted mode is never a default. It can only be switched on per chat, by users with the Warehouse remove permission (Lake remove permission for a Lake Builder). See Warehouse Builder and Lake Builder.
If the save is refused, the reason appears under the cards.
The model marked Default on the Mind. For a new Mind, that is Kartlos Premium.
The option appears only when Swarm is included for the Mind. Switch on Parallel sub-agents under Swarm on the AI Config tab, then click Save.
Yes. Raise Max Execution Time up to 200 seconds, then click Save.
It starts the feature switched on in every new chat. Users can still switch it off in the message box.
Pick another default under Model Selection, then click Save. Or ask an administrator to allow the model again in the company's AI configuration.
No. The Mind has no thinking-effort setting; effort is chosen per message in the chat, and only where the model supports it. See Models in a chat.
Up to the value of Max concurrent agents (1-25), which is between 1 and 25. The default is 10.
Only for vectorized training cards. It sets how many passages a search returns and how similar each one must be.
Yes. Use Test a Mind with benchmarks to run multiple-choice questions against different models. The Benchmarks tab needs Edit Mind.
The Edit Mind window's AI Config tab shows a smaller set of controls. Max Memory Usage and Similar search appear only on the Mind's AI Config tab, so change them there.
You need Edit Mind. Without it, the save is refused, and the reason appears under the cards.