Models and usage limits
Choose a different AI model, understand the cloud notice and keep an eye on the quota.
The answers in the chat are written by an AI language model. By default, this is Loki, the model that runs on the LOKI appliance in your organisation. If your administrator has also connected models from cloud providers, you can choose between them in the chat.
Choose a model
If more than one model is available to you, the input field shows a button with the name of the current model (Select model). On narrow screens, it shows only the provider's logo. If there is only one model, the button is not shown.
- Click the button.
- Search for a model in the Search model... field or select it from the list. The models are grouped by provider.
- Your choice applies from the next question onwards, including in your other chats.
If you switch models in the middle of a chat, the history shows a dividing line “Switched to” with the name of the new model at that point.
Back to Loki after reloading
LOKI does not save your choice of model permanently. If you reload the page, the local model Loki is selected again.
Effort
Some models also offer the Effort section with the levels Fast, Medium (default) and Thorough. More thorough usually means better but slower answers. The selected level is shown in small type on the model button.
Cloud models
If you choose a model from a cloud provider, the model button lights up, and the following notice appears above the input field:
Cloud model active — model name. Your messages and uploaded documents are sent to an external provider.
This means that your questions, the chat history so far, your attachments and the passages that LOKI draws from your documents go to this provider. Only use a cloud model if that is acceptable for the content of your chat. If in doubt, stay with the local model Loki.
You can hide the notice with ×. It appears again after you reload the page.
Compliance check
Your organisation can set rules that detect confidential details such as names, email addresses or account numbers before they leave LOKI, that is, before they go to a cloud model or to the web search. Depending on the rule, one of the following then happens:
- LOKI redacts automatically. A notice “Content redacted before sending” shows which types of details were replaced.
- LOKI blocks the call.
- LOKI asks you. A “Compliance check” card lists what was detected and, depending on the rule, offers these options:
| Button | Effect |
|---|---|
| Handle locally | Nothing is sent; LOKI switches to the local model. |
| Send redacted | The detected details are replaced, and the rest is sent. |
| Send as-is | The content is sent unchanged. |
| Cancel call | Nothing is sent. |
Your decision is recorded in the compliance log, which administrators can view.
Usage limits
For cloud providers, your administrator can set a quota, measured in tokens (pieces of text). The quota applies to the whole organisation together, not per person. The local model Loki has no limit.
If a limit is set up for the provider of the selected model, the input field shows Token usage with a percentage next to the model button. The colour shows the status: turquoise means fine, yellow means the limit will soon be reached, red means it is used up.
Click it to see details: usage for the day, the week and the month, the limit that applies in each case (Current limit) and when it is reset.
| Status | What happens |
|---|---|
| from 80 per cent | “Token quota (period) is almost used up.” appears above the input field. |
| 100 per cent | The next time you send, LOKI automatically switches to the local model and states the time from which cloud models are available again. An answer that reaches the limit is stopped at that point. |