ThinkingSettings
Settings for extended thinking in Anthropic models.
See:
- https://docs.aws.amazon.com/bedrock/latest/userguide/claude-messages-extended-thinking.html
- https://docs.litellm.ai/docs/providers/anthropic_effort
- https://docs.litellm.ai/docs/providers/anthropic#usage---thinking--reasoning_content
ThinkingSettings
info
ThinkingSettings is defined beneath the following ancestor nodes in the YAML structure:
Below are the properties for the ThinkingSettings. Each property links to the specific details section further down in this page.
| Property | Default | Type | Required | Description |
|---|---|---|---|---|
| enabled | True | boolean | No | Whether to enable extended thinking |
| budget_tokens | integer | Yes | Maximum number of tokens to allocate for thinking (minimum: 1024) |
Property Details
OttoAgent
Model for a custom Otto agent.
| Property | Default | Type | Required | Description |
|---|---|---|---|---|
| id | string | No | Unique identifier for the agent | |
| name | string | Yes | The name of the agent | |
| extends | string | No | Agent ID or name to inherit from | |
| instructions | string | Yes | The instructions for the agent | |
| mcp_servers | array[string] | No | The MCP servers to use for the agent | |
| model | Any of: string ProviderModel | No | The model to use for the agent | |
| model_settings | OttoModelSettings | No | The model settings for the agent | |
| tools | array[string] | No | The tools to use for the agent | |
| max_turns | integer | No | Maximum number of turns for the agent | |
| jinja | False | boolean | No | Whether to render with Jinja templating |
OttoModelSettings
Model for the model settings for an Otto agent.
| Property | Default | Type | Required | Description |
|---|---|---|---|---|
| reasoning | string | No | Reasoning effort level for reasoning models (e.g., o1, o3). For OpenAI models: 'minimal', 'low', 'medium', 'high'. For Anthropic models via LiteLLM, this enables extended thinking and sets thinking budget tokens. 'low', 'medium', 'high' maps to budget tokens: low=1024, medium=2048, high=4096. Note: Starting from Claude Opus 4.5, reasoning_effort is separate from thinking.budget_tokens. Both can be used together - reasoning_effort affects all tokens (text, tools, thinking), while thinking.budget_tokens sets a specific maximum for extended thinking tokens. | |
| thinking | ThinkingSettings | No | Extended thinking settings for Anthropic models. Configures the thinking.type and thinking.budget_tokens parameters. Only applicable to Anthropic models. | |
| temperature | number | No | Model temperature setting. For Anthropic models, must be set to either None or 1. |