Brain & Models
The Brain tab controls which AI model powers your agent's intelligence. Different models have different strengths.
Supported providers
| Provider | Models | Strengths |
|---|---|---|
| OpenAI | GPT-4o, GPT-4o-mini, GPT-4-turbo, GPT-3.5-turbo | Best all-around, strong tool use, fast |
| Anthropic | Claude 3.5 Sonnet, Claude 3 Haiku, Claude 3 Opus | Long context, nuanced reasoning, safe |
| Google Gemini | Gemini Pro, Gemini Flash | Multi-modal (images + text), fast |
| Mistral | Mistral Large, Mistral Small | Cost-effective, fast, good for simple tasks |
| TinyGPT | Default model | Built-in; no API key needed, works out of the box |
Configuration
| Setting | Description | Range |
|---|---|---|
| Provider | Which AI company's model to use | Select from dropdown |
| Model | Specific model version | Depends on provider |
| Temperature | Controls randomness/creativity | 0.0 (deterministic) to 1.0 (creative) |
| Max tokens | Maximum response length | Depends on model (typically 1K-128K) |
| API key | Your API key for the provider | Bring your own or use saved keys |
Azure OpenAI support
For enterprise users with Azure-hosted OpenAI:
| Setting | Description |
|---|---|
| Azure endpoint | Your Azure OpenAI endpoint URL |
| Deployment name | The model deployment name in Azure |
| API version | Azure API version string |
Choosing the right model
| Use case | Recommended model | Why |
|---|---|---|
| General support bot | GPT-4o-mini or TinyGPT | Fast, cost-effective, handles most queries |
| Complex reasoning | GPT-4o or Claude 3.5 Sonnet | Better at multi-step thinking |
| Long document Q&A | Claude 3.5 Sonnet | 200K token context window |
| Budget-conscious | Mistral Small or GPT-3.5-turbo | Cheapest per-token cost |
| Image understanding | Gemini Pro | Native multi-modal support |
| No API key setup | TinyGPT | Works immediately, no configuration |
Temperature guide
| Temperature | Behavior | Use when |
|---|---|---|
| 0.0 | Always picks the most likely response | Factual Q&A, classification, data extraction |
| 0.3 | Mostly consistent with slight variation | Customer support, documentation help |
| 0.7 | Balanced creativity and consistency | General conversation (default) |
| 1.0 | Maximum creativity and randomness | Creative writing, brainstorming |
Tip
Start with TinyGPT to test your agent quickly without API key setup. Once you're happy with the agent's behavior, switch to a provider-specific model for better quality or lower cost.
Warning
Each provider has different pricing. Using GPT-4o or Claude 3.5 Sonnet costs significantly more per interaction than GPT-3.5-turbo or Mistral Small. Check your credit usage regularly.
Note
API keys are stored securely and never exposed in the agent's public-facing chat. You can also save keys at the workspace level to reuse across multiple agents.