Model settings and image input
Choose providers and models, including vision-capable models for attached images.
Under Settings → AI model, Admins add named models with a provider, model id, API key and prices. Choose one when starting a task, or switch it beside the chat composer. A fast model can handle history summaries; the main model handles the task. Keys stay encrypted and are never shown to agents.
Image input applies to the main model:
- Auto enables known vision families. Unknown compatible-provider or local aliases stay text-only.
- Supported — send images enables a custom model or alias that accepts visual inputs.
- Not supported — text only supplies an explanatory notice instead of pixels.
The Images badge means image sending is enabled. Test connection tests a text reply only; it does not prove vision support. The selected model must also support tool calling. Mock provides scripted answers and does not interpret images.
The API accepts supports_images: true, false or null (Auto) on model profiles.
The server operator configures its own default with Llm:SupportsImages, and additional server
models with Llm:ServerModels:<index>:SupportsImages. Image count and size limits are separate
from file-upload limits. See Tasks for supported formats and usage.