Google Gemini AI Model Configuration in ConvyAI
Overview
The Google Gemini integration connects ConvyAI to Google's multimodal foundation models for intent classification, entity extraction, and response generation within the bot flow.
Prerequisites
- A Google Gemini API key.
- API key configured in Generative AI Settings.
Create a Gemini AI Model
Navigate to Generative AI → AI Models and click New Model → Gemini.
Configuration options
| Field | Description |
|---|---|
| Name | A unique name to identify this AI Model. |
| API Key | Your Google Gemini API key. |
ConvyAI supports Gemini models 3.5 and later accessible via the Google AI API. Gemini 2.5 models are deprecated by Google and can no longer be registered. The model is specified by name in the Completion model advanced parameter (for example gemini-3.8-flash). Refer to the Google AI model documentation for the full list of available models and their identifiers.
Advanced parameters
| Parameter | Default | Description |
|---|---|---|
| Prompt version | v3 | System prompt version applied when using the model in Knowledge Base nodes. |
| Completion model | gemini-3.8-flash | The Gemini model used for response generation. Free text: enter the model identifier as provided by Google. |
| Thinking Level | Empty | Reasoning level of the model. Free text, for example minimal, low, medium or high. |
| Max token completion | Empty | Maximum number of tokens in the generated response. Optional. |
| Number of extracted chunks | 3 | How many Knowledge Base chunks are retrieved and passed to the model as context during RAG. |
Optional parameters
Thinking Level and Max token completion are optional. If you leave a field empty, the parameter is not sent and the model uses its default. If you enter a value, it is sent to the provider as is, without checks: make sure the value is valid for your model. If the provider rejects it, the test console shows its error. See Optional parameters.
Configure Thinking Level once on the AI Model: the AI Agent nodes of both FlowDesigner and Agentic Designer use it.
Removed parameters
Thinking budget, Temperature, Top P, Frequency penalty and Presence penalty are no longer available on Gemini AI Models, because Gemini 3.5 and later models do not support them. They are not sent to the provider even if they are present in an AI Model saved earlier.
Upgrade from Gemini 2.5
AI Models already configured on Gemini 2.5 models stop working: the test console shows the error returned by the provider and flows do not crash. To restore them, enter a 3.5 or later model in Completion model and, if needed, set Thinking Level. AI Models already on 3.5 or later keep working without reconfiguration.
Intents, entities, and testing
Configuring intents and entities and testing the model are features shared across all AI Model providers. See Intents, Entities & Testing for full documentation.