Doubao Seed 2.0 Responses API - Complete API Reference

## All APIs require Bearer Token authentication ## **Get API Key:** Visit the [API Key Management Page](https://starmagic.ai/app/api-keys) to get your API Key **Add to request headers:** ``` Authorization: Bearer YOUR_API_KEY ```

Authorization: Bearer YOUR_API_KEY

Chat model name - `doubao-seed-2.0-pro`: Flagship, strongest overall capability - `doubao-seed-2.0-lite`: Lightweight, faster speed - `doubao-seed-2.0-mini`: Ultra-fast, quickest response - `doubao-seed-2.0-code`: Code-specialized

"doubao-seed-2.0-pro"

Input content. Supports two formats: **1. Plain text string**: Equivalent to text input from the user role **2. Element list (array)**: Contains multiple input types including messages, context, tool calls, etc.

Insert a system message or developer instruction as the first instruction in the model context **Note**: - When used with previous_response_id, instructions from the previous response are not inherited - Cannot be used together with caching

"string"

Unique identifier of the previous model response, used to implement multi-turn conversations **Note**: - When provided, it will include the input and response content from the previous round, and the input tokens for the current request will increase accordingly - In continuous multi-turn conversations, it is recommended to add approximately 100ms delay between each request

"string"

Storage expiration time, UTC Unix timestamp (seconds) **Note**: - Value range: (creation time, creation time + 604800], i.e., maximum retention of 7 days - Default: creation time + 259200 (3 days) - Applies to both store and caching - Cache storage is billed by time, less than 1 hour is counted as 1 hour

0

Maximum number of output tokens from the model, including model response and chain-of-thought content

32768

Controls whether the model enables deep thinking mode

{
  "type": "enabled"
}

Limit deep thinking workload, reducing it can make responses faster and use fewer tokens

{
  "effort": "medium"
}

Whether to enable context caching **Note**: Cannot be used together with instructions field or tools field (except Function Calling)

{
  "type": "disabled",
  "prefix": false
}

Whether to store the generated model response for later retrieval via API - `true`: Store the current model response - `false`: Do not store

true

Whether to stream the response content - `false`: Return all content at once after the model finishes generating - `true`: Return chunks progressively via SSE protocol, ending with a `data: [DONE]` message

false

Sampling temperature, controls output randomness - Value range: [0, 2] - Lower values are more deterministic, higher values are more random - It is recommended to adjust only one of temperature or top_p **Note**: doubao-seed-2.0-pro and doubao-seed-2.0-lite are fixed at 1, manual specification will be ignored

1

Nucleus sampling probability threshold - Value range: [0, 1] - It is recommended to adjust only one of temperature or top_p **Note**: doubao-seed-2.0-pro and doubao-seed-2.0-lite are fixed at 0.95, manual specification will be ignored

0.7

Format definition for model text output

{
  "format": {
    "type": "text",
    "name": "string",
    "schema": {},
    "description": "string",
    "strict": false
  }
}

List of tools the model can call Supported tool types: - `function`: Custom function (Function Calling) - `web_search`: Web search - `doubao_app`: Doubao App - `image_process`: Image processing - `mcp`: MCP tools - `knowledge_search`: Private knowledge base search

[
  {
    "type": "function",
    "name": "string",
    "description": "string",
    "parameters": {},
    "strict": true,
    "feature": {
      "chat": {
        "type": "disabled",
        "role_description": "string"
      },
      "deep_chat": {
        "type": "disabled",
        "role_description": "string"
      },
      "ai_search": {
        "type": "disabled",
        "role_description": "string"
      },
      "reasoning_search": {
        "type": "disabled",
        "role_description": "string"
      }
    },
    "user_location": {
      "type": "approximate",
      "country": "string",
      "region": "string",
      "city": "string"
    },
    "sources": [
      "toutiao"
    ],
    "limit": 0,
    "max_keyword": 0,
    "point": {
      "type": "enabled"
    },
    "grounding": {
      "type": "enabled"
    },
    "zoom": {
      "type": "enabled"
    },
    "rotate": {
      "type": "enabled"
    },
    "server_label": "string",
    "server_url": "string",
    "headers": {},
    "require_approval": null,
    "allowed_tools": null,
    "knowledge_resource_id": "string",
    "doc_filters": {},
    "dense_weight": 0.5,
    "ranking_options": {
      "rerank_switch": false,
      "retrieve_count": 25,
      "get_attachment_link": false,
      "chunk_diffusion_count": 0,
      "chunk_group": false,
      "rerank_model": "base-multilingual-rerank",
      "rerank_only_chunk": false
    }
  }
]

Controls whether the model calls tools **String mode**: - `none`: Do not call tools - `required`: Must call tools - `auto`: Model decides (default) **Object mode**: Specify a particular tool to call

Maximum number of tool call rounds (unlimited calls per round) - Value range: [1, 10] - Web Search default: 3 - Image Process default: 10 (modification not supported) - Knowledge Search default: 3 **Note**: This parameter operates on a best-effort basis; the actual number of calls is influenced by model inference performance and other factors

0

Context management strategy to help the model effectively utilize the context window

{
  "edits": [
    {
      "type": "clear_thinking",
      "keep": null,
      "exclude_tools": [
        "string"
      ],
      "clear_tool_input": false,
      "trigger": {
        "type": "tool_uses",
        "value": 0
      }
    }
  ]
}