AI Models - Configure LLM Providers for Generative AI in ConvyAI
What are AI Models
An AI Model in ConvyAI is a configured connection to a Large Language Model provider. It defines which LLM engine to use, the authentication credentials, and how the model should behave - including intent and entity extraction parameters.
AI Models are the shared LLM infrastructure used across ConvyAI's Generative AI capabilities:
- In the Knowledge Base node - to analyze user input, classify intent, extract entities, and generate RAG-grounded responses.
- In the AI Agent node - to orchestrate tool calls, process images, and produce outputs that drive flow logic.
Supported providers
| Provider | Models | Notes |
|---|---|---|
| OpenAI | GPT-3.5 and above | Requires OpenAI API key |
| Amazon Bedrock | Anthropic Claude (multiple versions) | Requires AWS credentials |
| Google Gemini | Gemini 3.5 and later | Requires Gemini API key |
| Azure OpenAI | Azure-deployed OpenAI models | Requires Azure endpoint and key |
| Xference | Qwen 3.6, Gemma 4 3B, Gemma 4 12B | Requires Xference endpoint and key; inference runs on Italian datacenters |
| Bring your own Model | Any model with a REST API | HTTP REST integration |
Optional parameters
Optional parameters of an AI Model (such as Temperature, Top P, penalties, reasoning effort, extended thinking budget, Thinking Level and Thinking Budget) follow the same rule for every provider and model:
- An empty field means the parameter is not sent, and the provider default applies.
- Any value you enter is sent to the provider as is.
ConvyAI does not disable, hide or force any parameter based on the model you enter. New AI Models start with these fields empty, and an information icon next to the fields reminds you of the rule. If a model does not accept a parameter, the provider rejects the call: use the test console to find out before deploying.
Reasoning settings (reasoning effort on OpenAI and Azure OpenAI, extended thinking budget on Amazon Bedrock, Thinking Level on Google Gemini, Thinking Budget on Xference) are configured once on the AI Model and apply to Knowledge Base nodes and AI Agent nodes alike, in both FlowDesigner and Agentic Designer.
AI Models created before version 7.9.0 keep their saved values and continue to send them.