API Gateway:Configure an AI log policy

Panel layout

Basic fields

The main basic fields are organized by category as follows:

JSON PATH configuration

JSON PATH configuration syntax

The JSON PATH configuration uses GJSON syntax to extract specified field values from the JSON data of an API request or response.

Syntax description:

• Use a period (.) to access a field in an object.

• Access array elements by index, with 0 being the first index.

• Multi-level nested access is supported.

For example, choices.0.message.content means:

• First, obtain the choices array.

• Then, retrieve the element at index 0.

• Then, sequentially retrieve the value of the message.content field.

Example:

The following is a sample response:

{
  "choices": [
    {
      "message": {
        "role": "assistant",
        "content": "I am a large-scale language model developed by Alibaba Cloud. My name is Qwen."
      },
      "finish_reason": "stop",
      "index": 0
    }
  ],
  "object": "chat.completion",
  "usage": {
    "prompt_tokens": 3019,
    "completion_tokens": 104,
    "total_tokens": 3123,
    "prompt_tokens_details": {
      "cached_tokens": 2048
    }
  },
  "model": "qwen-plus",
  "id": "chatcmpl-6ada9ed2-7f33-9de2-8bb0-78bd4035025a"
}

The following table shows sample JSON PATH configurations for the corresponding fields:

JSON PATH differences across paths and sources

A Model API may be associated with multiple paths, such as /v1/chat/completions, /v1/completions, and /v1/responses. The JSON PATH for the same field can differ depending on the path and source. The following table shows examples for three representative fields: question, answer, and input_token.

Batch add fields from a template

The AI request log feature includes built-in templates to add predefined fields in batches.

1. In the AI request log panel, click Batch Add from Template.

2. Select Select Template, select the fields to add, and click Batch Add.

There are 15 built-in templates, divided into the following two groups by provider:

OpenAI templates (6)

Model Studio templates (9)

Add a custom field

In addition to basic fields and template fields, you can also add custom fields manually.

1. In the AI request log panel, click Add Custom Field.

2. Configure the following parameters:

   Parameter	Description
   Field name	The custom field name is used as the Key within the ai_log field of the log.
   Source	The source of the field. Valid values: Request, non-streaming response, and streaming response.
   JSON PATH	The JSON PATH used to extract the field value from the source data.
   Description	A description of the field's purpose.

3. Click Confirm to complete the configuration.

After you add a custom field, it appears on the Custom Fields tab. Use Edit and Delete in the Actions column to manage existing fields.

Example 1: Log core fields for OpenAI-compatible protocols

This example shows how to log non-streaming conversation content and token usage for the /v1/chat/completions path to reduce log size.

1. Go to the Policies and Plug-ins tab of the target Model API and expand the AI Request Log panel.

2. Select the /v1/chat/completions path.

3. On the Basic Fields tab, enable the following fields:

   Field name	Source	Category	Purpose	Enabled by default
   question	Request	Conversation	Logs the user's question.	No
   answer	non-streaming response	Conversation	Logs the model's answer in a non-streaming response.	No
   input_token	non-streaming response	Usage	Logs the number of input tokens.	Yes
   output_token	non-streaming response	Usage	Logs the number of output tokens.	Yes
   total_token	non-streaming response	Usage	Logs the total number of tokens.	Yes

4. Click Save.

Example 2: Batch add Model Studio fields from template

This example shows how to quickly add log fields for the Model Studio text generation protocol using a built-in template.

1. Go to the Policies and Plug-ins tab of the target Model API and expand the AI Request Log panel.

2. Click Batch Add from Template.

3. Set Select Template to /api/v1/services/aigc/text-generation/generation.

4. Select the checkboxes for the fields that you need, and then click Batch Add.

5. Based on your requirements, enable or disable specific fields in the field list, or modify their JSON PATHs.

6. Click Save.