Model Assignments shows which LLM your AI agents and your Ask LLM, RAG Agent and Virtual Agent nodes use. They are grouped by model, so you can change a model across the whole project at once. You can use it to move a project to a new model, or off a model that is being retired. For an overview, see Projects.
To open it, click the gear icon on the project card, then click Model Assignments in the Project Settings sidebar. To view the page you need access to projects and to LLM configurations. To save, you need permission to edit projects.
What the page lists
| Component | What is included |
|---|---|
| AI Agents | The AI agents you created. Agent Builder, Improver and Optimizer are not listed. |
| Ask LLM, RAG Agent and Virtual Agent nodes | Nodes from every flow in the project: the main flow, subflows, survey flows, the Post Session Flow, AI agent tool flows and virtual agent intent flows. |
Guardrails, knowledge bases, AI Insights and Add-On AI Features are not listed. They keep their own LLM settings, so you change them where they are configured.
If no component in the project uses an LLM, the page shows "No model assignments found for this project."
The page
1. Draft and Published: You can edit the page only on the draft. On the published version the fields are disabled and there is no save button.
2. Group header: There is one card for each group. A group is the components that use the same LLM configuration, model and reasoning effort, so one model at two reasoning efforts makes two cards. The header shows the model (for example gemini-3.1-flash-lite), the LLM configuration name and its provider (Gemini, Google Gemini).
3. Members: The components in the group, in rows: AI Agents, Ask LLM Nodes, RAG Agent Nodes and Virtual Agent Nodes. A row appears only if the group has components of that kind.
4. LLM Configuration, Model and Reasoning Effort: The settings that every component in the card uses. LLM Configuration and Model are required.
5. Save: The page has one save button, at the bottom right.
Changing a model
In this example, the three AI agents and two RAG Agent nodes that use gemini-3.1-flash-lite are moved to gemini-3.5-flash.
- Open the project's Draft.
- Find the card for the model you are replacing, here
gemini-3.1-flash-lite. - Change the fields. Every component in the card moves together, and you cannot choose individual agents or nodes here. To change a single component, change it on that AI agent or node instead. A changed card is marked Modified.
- LLM Configuration: Changing it clears Model. If the new configuration offers only one model, that model is selected for you.
- Model: Lists the configuration's suggested models. In this example, choose
gemini-3.5-flash. To use a model that is not listed, choose custom model (the last option) and type the name in the Model Name field that appears. An Azure OpenAI configuration has no model list. - Reasoning Effort: Default, None, Low, Medium or High. The field appears only for models that support reasoning. If you choose a model without reasoning, the field is cleared.
- If other cards need changes too, change them now. A single save covers all changed cards.
- Save. The button you see depends on what you changed:
| Button | Shown when | What it does |
|---|---|---|
| Save & Prepare Agents (N) | A changed card contains AI agents. N is the number of AI agents in the changed cards. | Saves, then prepares those AI agents in the same step. |
| Save (in the Save & Prepare Agents dropdown) | A changed card contains AI agents. | Saves only. The changed AI agents use the new model only after you prepare them again. See Creating an AI Agent. |
| Save | Only nodes changed. | Saves. Nodes do not need to be prepared. |
In this example the card holds three AI agents, so the button reads Save & Prepare Agents (3).
- Publish the project. Your change is saved in the draft, and live conversations start using it only after you publish.
💡 Tip: After you save, check the preparation status of each changed AI agent on the AI Agents page. An agent can fail to prepare with the new model even when the save succeeds, for example when the model does not accept the reasoning effort you chose. If an agent failed, prepare it from its own page to see the reason.