ai-request-rewrite

Description

The ai-request-rewrite plugin intercepts client requests before they are forwarded to the upstream service. It sends a predefined prompt, along with the original request body, to a specified LLM service. The LLM processes the input and returns a modified request body, which is then used for the upstream request. This allows dynamic transformation of API requests based on AI-generated content.

Plugin Attributes

Field	Required	Type	Description
prompt	Yes	String	The prompt send to LLM service.
provider	Yes	String	Name of the LLM service. Available options: openai, deekseek and openai-compatible
auth	Yes	Object	Authentication configuration
auth.header	No	Object	Authentication headers. Key must match pattern ^[a-zA-Z0-9._-]+$.
auth.query	No	Object	Authentication query parameters. Key must match pattern ^[a-zA-Z0-9._-]+$.
options	No	Object	Key/value settings for the model
options.model	No	String	Model to execute. Examples: "gpt-3.5-turbo" for openai, "deepseek-chat" for deekseek, or "qwen-turbo" for openai-compatible services
override.endpoint	No	String	Override the default endpoint when using OpenAI-compatible services (e.g., self-hosted models or third-party LLM services). When the provider is 'openai-compatible', the endpoint field is required.
timeout	No	Integer	Total timeout in milliseconds for requests to LLM service, including connect, send, and read timeouts. Range: 1 - 60000. Default: 30000
keepalive	No	Boolean	Enable keepalive for requests to LLM service. Default: true
keepalive_timeout	No	Integer	Keepalive timeout in milliseconds for requests to LLM service. Minimum: 1000. Default: 60000
keepalive_pool	No	Integer	Keepalive pool size for requests to LLM service. Minimum: 1. Default: 30
ssl_verify	No	Boolean	SSL verification for requests to LLM service. Default: true

How it works

Examples

The examples below demonstrate how you can configure ai-request-rewrite for different scenarios.

note

You can fetch the admin_key from config.yaml and save to an environment variable with the following command:

admin_key=$(yq '.deployment.admin.admin_key[0].key' conf/config.yaml | sed 's/"//g')

Redact sensitive information

curl "http://127.0.0.1:9180/apisix/admin/routes/1" -X PUT \
  -H "X-API-KEY: ${admin_key}" \
  -d '{
    "uri": "/anything",
    "plugins": {
      "ai-request-rewrite": {
        "prompt": "Given a JSON request body, identify and mask any sensitive information such as credit card numbers, social security numbers, and personal identification numbers (e.g., passport or driver'\''s license numbers). Replace detected sensitive values with a masked format (e.g., \"*** **** **** 1234\") for credit card numbers. Ensure the JSON structure remains unchanged.",
        "provider": "openai",
        "auth": {
          "header": {
            "Authorization": "Bearer <some-token>"
          }
        },
        "options": {
          "model": "gpt-4"
        }
      }
    },
    "upstream": {
      "type": "roundrobin",
      "nodes": {
        "httpbin.org:80": 1
      }
    }
  }'

Now send a request:

curl "http://127.0.0.1:9080/anything" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "John Doe",
    "email": "john.doe@example.com",
    "credit_card": "4111 1111 1111 1111",
    "ssn": "123-45-6789",
    "address": "123 Main St"
  }'

The request body send to the LLM Service is as follows:

{
  "messages": [
     {
       "role": "system",
       "content": "Given a JSON request body, identify and mask any sensitive information such as credit card numbers, social security numbers, and personal identification numbers (e.g., passport or driver's license numbers). Replace detected sensitive values with a masked format (e.g., '*** **** **** 1234') for credit card numbers). Ensure the JSON structure remains unchanged."
     },
     {
       "role": "user",
       "content": "{\n\"name\":\"John Doe\",\n\"email\":\"john.doe@example.com\",\n\"credit_card\":\"4111 1111 1111 1111\",\n\"ssn\":\"123-45-6789\",\n\"address\":\"123 Main St\"\n}"
     }
   ]
}

The LLM processes the input and returns a modified request body, which replace detected sensitive values with a masked format then used for the upstream request:

{
  "name": "John Doe",
  "email": "john.doe@example.com",
  "credit_card": "**** **** **** 1111",
  "ssn": "***-**-6789",
  "address": "123 Main St"
}

Send request to an OpenAI compatible LLM

Create a route with the ai-request-rewrite plugin with provider set to openai-compatible and the endpoint of the model set to override.endpoint like so:

curl "http://127.0.0.1:9180/apisix/admin/routes/1" -X PUT \
  -H "X-API-KEY: ${admin_key}" \
  -d '{
    "uri": "/anything",
    "plugins": {
      "ai-request-rewrite": {
        "prompt": "Given a JSON request body, identify and mask any sensitive information such as credit card numbers, social security numbers, and personal identification numbers (e.g., passport or driver'\''s license numbers). Replace detected sensitive values with a masked format (e.g., '*** **** **** 1234') for credit card numbers). Ensure the JSON structure remains unchanged.",
        "provider": "openai-compatible",
        "auth": {
          "header": {
            "Authorization": "Bearer <some-token>"
          }
        },
        "options": {
          "model": "qwen-plus",
          "max_tokens": 1024,
          "temperature": 1
        },
        "override": {
          "endpoint": "https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions"
        }
      }
    },
    "upstream": {
      "type": "roundrobin",
      "nodes": {
        "httpbin.org:80": 1
      }
    }
  }'