LLMs

Illinois Chat is model-agnostic: each chatbot chooses which large language models power it, and users can switch models per conversation. Everything on this page is set on the chatbot's LLMs screen ("Configure LLM Providers for your Chatbot").

Supported providers

The page groups providers in two sets:

Group	Providers
Closed models (bring your own key)	Anthropic, OpenAI, OpenAI Compatible (any endpoint that speaks the OpenAI API, such as vLLM — base URL including /v1), Azure OpenAI, Amazon Bedrock, Google Gemini, SambaNova
Open models (no key needed)	NCSA Hosted LLMs, NCSA Hosted VLMs, Ollama, WebLLM (runs entirely in the user's browser)

Free NCSA-hosted models are a zero-cost starting point; commercial providers generally give better instruction-following and citation quality, and you pay the provider directly for usage.

Bring your own key

For commercial providers you paste your own API key on the LLMs page. Keys are stored encrypted per chatbot and used to serve the chatbot's requests from both the chat interface and the Chat API. Your data is never used to train models — provider interactions are contractually protected.

Enabling models

• Each provider card has a switch; inside it, each model has its own switch. Users and the API can only use models that are enabled.
• Default Model picks the model new conversations start with. Users can choose a different enabled model from the model picker in chat; API callers pass model.
• Models hosted in countries of concern show a warning before you enable them or set them as default (Enable anyway / Set as default anyway).

Temperature is not set on this page: users choose it per conversation in the chat Settings › Model tab, and it defaults to 0.1.

Vision and tools

• Image input is supported on vision-capable models (for example GPT-4o or Claude).
• Tool selection does not use the chat model you pick here; see Tool routing.

The embedding model used for retrieval is set by the operator, not per chatbot — see the Configuration reference.