Using the AI Request Transformer Plugin

このページは、まだ日本語ではご利用いただけません。翻訳中です。

古いプラグインバージョンのドキュメントを閲覧しています。

Overview

The AI Request Transformer plugin is designed to operate in two ways:

• As a transformer/security arbiter for your existing upstream APIs
• As an extension of another AI Proxy LLM route, inspecting and transforming the requests before sending to the upstream LLM service

The plugin configuration consists of two distinct sections:

• The LLM configuration, which uses the same configuration options as the AI Proxy plugin.
• The prompt (and additional options) containing the instructions for the LLM, which will transform your request.

See the same LLM block in the context of the AI Proxy plugin and the AI Request Transformer plugin:

config:
  route_type: "llm/v1/chat"
  auth:
    header_name: "Authorization"
    header_value: "Bearer <openai_key>"  # add your own OpenAI API key
  model:
    provider: "openai"
    name: "gpt-4"

llm:
  config:
    route_type: "llm/v1/chat"
    auth:
      header_name: "Authorization"
      header_value: "Bearer <openai_key>"  # add your own OpenAI API key
    model:
      provider: "openai"
      name: "gpt-4"

prompt: "Transform my message to XML - return me ONLY the XML output."

When the plugin is accessed in any scope (global, service, route, or consumer), it always sets the caller’s request body as the user prompt in a chat message, and then sends it to the configured llm: configuration block.

Examples

Transforming existing API traffic

This example uses AI Request Transformer on an existing API, for example, something that you have already developed and maintain internally.

1. Design the prompt.

   For this example, we want to intercept requests to our customers API. On each request, we want to first forward the request to our configured large language model and ask the LLM to add the country name field to anywhere in the JSON where there is a city, but no associated country.

   The plugin would be configured like this:

    config:
      prompt: >
        In my JSON message, anywhere there is a JSON tag for a "city", also add a "country" tag with the name of the country in which the city
        resides. Return me only the JSON message, no extra text.
      llm:
        # see `ai-proxy` plugin documentation for compatible fields for the "llm" block
   

2. Attach the plugin.

   Attach the ai-request-transformer plugin to the global level, route, service, or consumer on which you want to inspect or transform all requests.

3. What happens next?

   First, a Kong client makes a request. For example:

    {
      "user": {
        "name": "Kong User",
        "city": "London"
      }
    }
   

   Next, Kong Gateway parses this into an llm/v1/chat type message, based on your config.prompt:

    {
      "messages": [
        {
          "role": "system",
          "content": "In my JSON message, anywhere there is a JSON tag for a \"city\" also add a \"country\" tag with the name of the country in which the city resides. Only return the JSON message, no extra text."
        },
        {
          "role": "user",
          "content": "{\n\"user\":{\n\"name\":\"Kong User\",\n\"city\":\"London\"\n}\n}"
        }
      ]
    }
   

   Finally, it sends this to the configured LLM. On the response, it takes the trailing assistant response back from the LLM, and sets it as the outgoing HTTP body:

    {
      "user": {
        "name": "Kong User",
        "city": "London",
        "country": "United Kingdom"
      }
    }
   

Extraction patterns

If your LLM is a chatbot type, or is unpredictable in responses, you can configure the additional field transformation_extract_pattern with a (PCRE) regular expression to extract the first match from the LLM’s response.

For example, if you have asked for a JSON response but you know that your LLM may add its own text around your answer, use this extraction pattern to withdraw only the JSON object from the LLM’s response:

config:
  prompt: >
    In my JSON message, anywhere there is a JSON tag for a "city" also add a "country" tag with the name of the country in which the city
    resides. Only return the JSON message, no extra text.
  transformation_extract_pattern: '\\{((.|\n)*)\\}'