# Choose the chat model

The chat model is the model that answers users in a Web Agent's widget. It reads the persona, decides which actions to run and fills in their parameters. Each agent has its own chat model, set under **AI Model** in the agent's **Assistant** tab.

## The chat model and the model provider

Two settings have similar names and do different jobs.

| Setting | Where | What it decides |
| --- | --- | --- |
| Model provider | **Settings** > **Model Provider**, once for the workspace | Whose account every AI request runs on and is billed to: the analysis, the editors and the chat. |
| Chat model | The agent's **Assistant** tab > **AI Model**, once for each agent | Which of that provider's models answers in this agent's widget. |

The **Provider** cards under **AI Model** are **Anthropic** and **OpenAI**. They choose the family of models to list. They do not change the workspace's model provider. To set up or change the account itself, see [Model providers](https://docs.faustine.ai/workspace/model-providers/).

## Before you begin

Your workspace has a working model provider. The list of models comes from it.

## Choose the model

To change an agent's chat model:

1. In the portal, go to [**Web Agents**](https://auth.faustine.ai/go?to=%2Fweb-assistants) and select your agent.
2. Select the **Assistant** tab, then **Provider & model**.
3. In the **Provider** row, select **Anthropic** or **OpenAI**.
4. In the **Model** row, select a model. Each entry shows the model's ID and its context window. The entry marked **Default** is the platform default.
5. In the save bar, select **Save changes**.

The change applies to new messages once it is saved. The chat reads the model from the agent's settings, not from the version, so it does not wait for a publish and it applies to every environment.

To go back to the platform default, select the entry marked **Default**. An agent needs a model. With none selected, a conversation fails with a message that no chat model is configured.

## Which model to choose

The list has only models that Faustine has vetted for agents. Among them, a more capable model follows the persona more closely, picks the right action more often and fills in parameters correctly. A smaller model answers faster and costs less.

[Web search](https://docs.faustine.ai/web-agents/web-search/) works with Anthropic models only.

## Troubleshooting

| Symptom | Cause | Fix |
| --- | --- | --- |
| **No OpenAI models mapped — map one in Settings to offer it for chat** | Your model provider maps models by hand, and no OpenAI model is mapped. | Map one under **Settings** > **Model Provider**, or select **Anthropic**. |
| **No chat models mapped for your provider** | Your model provider has no chat model mapped. | Map the models under **Settings** > **Model Provider**. |
| The row shows a message that no API key is configured for the provider | Your workspace's model provider has no key for that family of models. | Select the other provider card. An Anthropic key is part of your model provider under **Settings** > **Model Provider**. To use OpenAI models, contact Faustine. |
| The selected model has the badge **Current · not vetted** | The model was selected before it left the vetted list. | Select a vetted model when you next change the setting. |

## Next steps

- [Model providers](https://docs.faustine.ai/workspace/model-providers/): the account the chat model runs on.
- [Write the agent's persona](https://docs.faustine.ai/web-agents/persona/): the instructions the model reads.
- [Turn on web search](https://docs.faustine.ai/web-agents/web-search/): available with Anthropic models.