---
title: "RAG Reasoning"
slug: "clovastudio-ragreasoning"
updated: 2026-10-08T04:23:36Z
published: 2026-10-08T04:23:36Z
canonical: "api.ncloud-docs.com/clovastudio-ragreasoning"
---

> ## Documentation Index
> Fetch the complete documentation index at: https://api.ncloud-docs.com/llms.txt
> Use this file to discover all available pages before exploring further.

# RAG Reasoning

Available in Classic and VPC

Generate evidence-based RAG answers by leveraging RAG Reasoning model trained on answer types, such as citation sources that increase credibility and citation source indexing notations. RAG Reasoning calls the engine in the function calling format. You can specify a single or multiple RAG functions, and LLM will autonomously select the best function for the context to generate search augmentation. When used in conjunction with [Reranker](/docs/en/clovastudio-reranker), you can get more stable results.

## Request

This section describes the request format. The method and URI are as follows:

| Method | URI |
| --- | --- |
| POST | /v1/api-tools/rag-reasoning |

### Request headers

For information about the request headers common to all CLOVA Studio APIs, see [CLOVA Studio request headers](/docs/en/ai-naver-clovastudio-summary#%EC%9A%94%EC%B2%AD%ED%97%A4%EB%8D%94).

### Request body

You can include the following data in the body of your request:

| Field | Type | Required | Description |
| --- | --- | --- | --- |
| `messages` | Array | Required | Conversation message: [messages](/docs/en/clovastudio-ragreasoning#messages) |
| `topP` | Double | Optional | Sample generated token candidates based on cumulative probability. - 0.00 ＜ `topP` ≤ 1.00 (default: 0.8) |
| `topK` | Integer | Optional | Sample K high-probability candidates from the pool of generated token candidates. - 0 ≤ `topK` ≤ 128 (default: 0) |
| `maxTokens` | Integer | Optional | Maximum number of generated tokens - 1024 ≤ `maxTokens` ＜ 4096 (default: 1024) |
| `temperature` | Double | Optional | Degree of diversity for the generated tokens (higher values generate more diverse sentences) - 0.00 ＜ `temperature` ≤ 1.00 (default: 0.50) |
| `repetitionPenalty` | Double | Optional | Degree of penalty for generating the same token (the higher the setting, the less likely it is to generate the same result repeatedly) - 0.0 ＜ `repetitionPenalty` ≤ 2.0 (default: 1.1) |
| `stop` | Array | Optional | Token generation stop character - [] (default) |
| `seed` | Integer | Optional | Adjust consistency level of output on model iterations. - 0: Randomize consistency level (default). - 1 ≤ `seed` ≤ 4294967295: `seed` value of result value you want to generate consistently, or a user-specified `seed` value |
| `includeAiFilters` | Boolean | Optional | Whether to display the AI filter results (degree of the generated results in categories such as profanity, degradation/discrimination/hate, sexual harassment/obscenity, etc.) - `true` (default) \| `false` - `true`: Display - `false`: Not display |
| `tools` | Array | Required | List of tools available for `Function calling`: [tools](/docs/en/clovastudio-ragreasoning#tools) |
| `toolChoice` | String \| Object | Optional | `Function calling` tool call behavior - `auto` : Model automatically calls tool (string). - `none` : Model generates normal answer without calling tool (string). - Model forces specific tool call (object). |
| `toolChoice.type` | String | Optional | Type of tool the `Function calling` model will call |
| `toolChoice.function` | Object | Optional | Tool to be called by the `Function calling` model - `function` (valid value) |
| `toolChoice.function.name` | String | Optional | Tool name to be called by the `Function calling` model |

              Note

              

The maximum number of input tokens for RAG Reasoning is 128,000, and the maximum number of output tokens is 4,096.

#### `messages`

The following describes `messages`.

| Field | Type | Required | Description |
| --- | --- | --- | --- |
| `role` | Enum | Required | Role of conversation messages - `system` \| `user` \| `assistant` \| `tool` - `system`: Directives that define roles - `user`: User utterances/questions - `assistant`: Answers to user utterances/questions - `tool`: Results of the function called by the assistant (model) |
| `content` | String | Required | Conversation message content - Enter text (string). |
| `toolCalls` | Array | Conditional | Assistant call tool information - If the `role` is a `tool`, enter as the assistant's [toolCalls](/docs/en/clovastudio-ragreasoning#toolCalls) request. |
| `toolCallId` | String | Conditional | Tool ID - Required if `role` is `tool`. - Used to connect to the assistant's [toolCalls](/docs/en/clovastudio-ragreasoning#toolCalls) request. |

              Note

              

If the `role` is `tool`, the `content` of `messages` should include the list of documents retrieved from the search database or search API (`search_result`). Include `id: {unique ID of the document}`, `doc: {original document retrieved}` in the search result (`search_result`) so that it can be used for citation marks in RAG answers. See below for an example.

```
{
    "role": "tool",
    "content": "[
                    {
                        \"search_result\": [{\"id\": \"doc-1493058999\",
                        \"doc\": \"Login with NAVER ID is only available for individual members. Business members can't use this feature.\"
                    },
                    ...
                 ]"
}
```

#### `tools`

The following describes `tools`.

| Field | Type | Required | Description |
| --- | --- | --- | --- |
| `type` | String | Required | Tool type - `function` (valid value) |
| `function` | Object | Required | Call `function` information |
| `function.name` | String | Required | `function` name |
| `function.description` | String | Required | `function` description |
| `function.parameters` | Object | Required | Parameter passed when using `function` - `properties`, `required` - Entry: See [Operation method](/beta/docs/en/clovastudio-chatcompletionsv3-fc#%EB%8F%99%EC%9E%91-%EB%B0%A9%EC%8B%9D). - Format: See [JSON Schema reference](https://json-schema.org/understanding-json-schema/reference). |

#### `toolCalls`

The following describes `toolCalls`.

| Field | Type | Required | Description |
| --- | --- | --- | --- |
| `id` | String | - | Tool identifier |
| `type` | String | - | Tool type - `function` (valid value) |
| `function` | Object | - | Call `function` information |
| `function.name` | String | - | `function` name |
| `function.arguments` | Object | - | Parameter passed when using `function` |

### Request example

The request example is as follows:

- Step 1. Enter your query in `role: user` and call the best function to generate the answer ([Check response](/docs/en/clovastudio-ragreasoning#%EC%9D%91%EB%8B%B5%EC%98%88%EC%8B%9C1))

```
curl --location --request POST 'https://clovastudio.stream.ntruss.com/v1/api-tools/rag-reasoning' \
--header 'Authorization: Bearer <access_token>' \
--header 'Content-Type: application/json' \
--data-raw '{
  "messages": [
    {
      "content": "How to rent an A100 GPU",
      "role": "user"
    }
  ],
  "tools": [
    {
      "function": {
        "description": "This is the tool you use to do Ncloud-related searches.\nUse the tool by breaking up your query if you need to ask multiple questions.\nIf you can't find information, you can use the tool again with suggested_queries as a reference without giving a final answer.",
        "name": "ncloud_cs_retrieval",
        "parameters": {
          "properties": {
            "query": {
              "description": "Refine and enter the user's search keywords.",
              "type": "string"
            }
          },
          "required": [
            "query"
          ],
          "type": "object"
        }
      },
      "type": "function"
    }
  ],
  "toolChoice": "auto",
  "maxTokens": 1024
}'
```
- Step 2. Request with `role: tool` to generate the final answer ([Check response](/docs/en/clovastudio-ragreasoning#%EC%9D%91%EB%8B%B5%EC%98%88%EC%8B%9C2)) 

```
curl --location --request POST 'https://clovastudio.stream.ntruss.com/v1/api-tools/rag-reasoning' \
--header 'Authorization: Bearer <access_token>' \
--header 'Content-Type: application/json' \
--data-raw '{
    "messages": [
        {
            "content": "How to rent an A100 GPU",
            "role": "user"
        },
        {
            "role": "assistant",
            "content": "",
            "toolCalls": [
                {
                    "id": "call_enTEYb0kWBjOwtkngbl7FGTm",
                    "type": "function",
                    "function": {
                        "name": "ncloud_cs_retrieval",
                        "arguments": {
                            "query": "How to rent an A100 GPU"
                        }
                    }
                }
            ]
        },
        {
            "content": "{\"search_result\": [{\"id\": \"doc-179\", \"doc\": \"GPU A100 can only be created in KR-1. When creating an A100, select a subnet in KR-1. Up to 5 GPU servers can be created for corporate members only.\"}, {\"id\": \"doc-248\", \"doc\": \"GPU A100 servers can be created in the Services > Compute > Server menu. For more information, see the Create server guide.\"}, {\"id\": \"doc-156\", \"doc\": \"For individual members who need more GPU servers or need to create a GPU server, please refer to the FAQ and contact Support.\"}]}",
            "role": "tool",
            "toolCallId": "call_enTEYb0kWBjOwtkngbl7FGTm"
        }
    ],
    "tools": [
        {
            "function": {
                "description": "This is the tool you use to do Ncloud-related searches.\nUse the tool by breaking up your query if you need to ask multiple questions.\nIf you can't find information, you can use the tool again with suggested_queries as a reference without giving a final answer.",
                "name": "ncloud_cs_retrieval",
                "parameters": {
                    "properties": {
                        "query": {
                            "description": "Refine and enter the user's search keywords.",
                            "type": "string"
                        }
                    },
                    "required": [
                        "query"
                    ],
                    "type": "object"
                }
            },
            "type": "function"
        }
    ]
}'
```

## Response

This section describes the response format.

### Response headers

The following describes the response headers.

| Headers | Required | Description |
| --- | --- | --- |
| `Content-Type` | - | Response data format - `application/json` |

### Response body

The response body includes the following data:

| Field | Type | Required | Description |
| --- | --- | --- | --- |
| `status` | Object | - | See [Response status](/docs/en/ai-naver-clovastudio-summary#%EC%9D%91%EB%8B%B5). |
| `result` | Object | - | Response result |
| `result.message` | ChatMessage | - | Conversation message list |
| `result.message.role` | Enum | - | Role of conversation messages - `system` \| `user` \| `assistant` - `system`: Directives that define roles - `user`: User utterances/questions - `assistant`: Answers of the model |
| `result.message.content` | String | - | Content of conversation messages |
| `result.message.thinkingContent` | String | - | Decision flow in model |
| `result.message.toolCalls` | Array | - | [toolCalls](/docs/en/clovastudio-ragreasoning#toolCalls_response) |
| `result.usage` | Object | - | Token usage |
| `result.usage.completionTokens` | Integer | - | Generated token count |
| `result.usage.promptTokens` | Integer | - | Number of input (prompt) tokens |
| `result.usage.totalTokens` | Integer | - | Total number of tokens - Number of generated tokens + number of input tokens |

#### `toolCalls`

The following describes `toolCalls`.

| Field | Type | Required | Description |
| --- | --- | --- | --- |
| `id` | String | - | Tool identifier |
| `type` | String | - | Tool type - `function` (valid value) |
| `function` | Object | - | Call `function` information |
| `function.name` | String | - | `function` name |
| `function.arguments` | Object | - | Parameter passed when using `function` |

### Response example

The response example is as follows:

- Example response to Step 1. ([Check response](/docs/en/clovastudio-ragreasoning#%EC%9A%94%EC%B2%AD%EC%98%88%EC%8B%9C1))

```
{
    "status": {
        "code": "20000",
        "message": "OK"
    },
    "result": {
        "message": {
            "role": "assistant",
            "content": "",
            "thinkingContent": "The user has inquired about \"how to rent an A100 GPU\". To find the answer to this question, you need to use the tool "ncloud_cs_retrieval" to retrieve relevant information.",
            "toolCalls": [
                {
                    "id": "call_enTEYb0kWBjOwtkngbl7FGTm",
                    "type": "function",
                    "function": {
                        "name": "ncloud_cs_retrieval",
                        "arguments": {
                            "query": "How to rent an A100 GPU"
                        }
                    }
                }
            ]
        },
        "usage": {
            "promptTokens": 135,
            "completionTokens": 84,
            "totalTokens": 219
        }
    }
}
```
- Example response to Step 2. (LLM returns the final answer) ([Check response](/docs/en/clovastudio-ragreasoning#%EC%9A%94%EC%B2%AD%EC%98%88%EC%8B%9C2))

```
{
    "status": {
        "code": "20000",
        "message": "OK"
    },
    "result": {
        "message": {
            "role": "assistant",
            "content": "To rent an A100 GPU, <doc-248>you can create a GPU A100 server from the Services > Compute > Server menu in the NAVER Cloud Platform console.</doc-248>However, <doc-179>GPU A100 can only be created in KR-1, and you must select a subnet in KR-1 when creating an A100.</doc-179> Also, <doc-179>up to GPU servers can be created for corporate members only.</doc-179> If you need more GPU servers or are an individual member who needs to create a GPU server, <doc-156>please refer to the FAQ and contact Support.</doc-156>"
        },
        "usage": {
            "promptTokens": 332,
            "completionTokens": 146,
            "totalTokens": 478
        }
    }
}
```

#### Failure

The following is a sample response upon a failed call.

- [Common client error troubleshooting (4xx)](/beta/docs/en/clovastudio-troubleshoot-c4xx)
- [Common server error troubleshooting (5xx)](/beta/docs/en/clovastudio-troubleshoot-c5xx)