Model Configuration

The **Model Configuration** panel lets you assign AI models to different roles and add new model providers — all through the GUI, no file editing required.

![Settings — Model Configuration](https://assets.yaoagents.com/en/settings/03-config-models-en-$__THEME.png)

## Model roles

Yao Agents uses a role-based model system. Each role serves a different purpose, and you can assign a different model (or provider) to each role:

| Role | Purpose | When it's used |
|------|---------|----------------|
| **Default** | General-purpose tasks | Most conversations, everyday Q&A, content generation. |
| **Heavy** | Complex reasoning | Deep analysis, long planning, multi-step problem solving. |
| **Light** | Fast, simple tasks | Quick replies, classification, short summaries. |
| **Vision** | Image understanding | Reading screenshots, analyzing charts, OCR tasks. |
| **Audio** | Speech-to-text | Voice input transcription. |
| **Embedding** | Text embedding | Semantic search, document similarity. |

## Assigning models to roles

1. Open **Settings → Model Configuration**.
2. For each role, select a model from the dropdown. Available models come from your configured providers.
3. Click **Save**.

![Settings — Model role assignment](https://assets.yaoagents.com/en/settings/04-config-models-add-en-$__THEME.png)

If you haven't added any providers yet, use the [Cloud Service](/docs/en-us/settings/cloud-service) — it gives you instant access to all roles without additional setup.

## Adding a model provider

To use models from OpenAI, Anthropic, DeepSeek, Google, or any OpenAI-compatible service:

1. Click **Add Provider** at the bottom of the Model Configuration page.
2. Select the provider type (OpenAI, Anthropic, DeepSeek, Google, Ollama, or Custom).
3. Enter the required information:

| Field | Description |
|-------|-------------|
| Name | A display label for this provider (e.g., "My OpenAI"). |
| API Host | The base URL of the provider's API. Pre-filled for known providers. |
| API Key | Your API key from the provider. |
| Models | The specific model identifiers to make available (e.g., `gpt-4o`, `claude-sonnet-4-5-20250929`). |

4. Click **Save**. The new models appear in the role dropdowns immediately.

![Settings — Add model provider](https://assets.yaoagents.com/en/settings/05-config-models-add-deepseek-en-$__THEME.png)

## Editing or removing a provider

Click the provider card to edit its settings. Use the delete button to remove a provider. Removing a provider unassigns any roles that were using its models — you'll need to reassign those roles.

## Supported providers

| Provider | Type | Notes |
|----------|------|-------|
| OpenAI | `openai` | GPT-4o, GPT-4.1, o3, o4-mini, etc. |
| Anthropic | `anthropic` | Claude Opus 4, Claude Sonnet 4.5, etc. |
| DeepSeek | `openai` | DeepSeek V4, R1, etc. OpenAI-compatible endpoint. |
| Google | `openai` | Gemini models via OpenAI-compatible gateway. |
| Ollama | `openai` | Local models. Set host to `http://127.0.0.1:11434/v1/`. |
| Any OpenAI-compatible | `openai` | vLLM, LiteLLM, Azure OpenAI, Groq, Together, etc. |

## Advanced: manual connector files

For advanced use cases (custom capabilities, thinking mode configuration, Azure deployments), you can still create connector files manually. See [Environment Variables](/docs/en-us/settings/env-vars) for the underlying `.conn.yao` file format and directory structure.