AI Proxy

The AI Proxy plugin lets you transform and proxy requests to a number of AI providers and models.

AI Proxy plugin accepts requests in one of a few defined and standardized OpenAI formats, translates them to the configured target format, and then transforms the response back into a standard format.

v3.10+ To use AI Proxy with non-OpenAI format without conversion, see section below for more details.

AI Proxy plugin supports capabilities across batch processing, multimodal embeddings, agents, audio, image, streaming, and more, spanning multiple providers.

For Kong Gateway versions 3.6 or earlier:

• Chat Completions APIs: Multi-turn conversations with system/user/assistant roles.

• Completions API: Generates free-form text from a prompt.

  OpenAI has marked this endpoint as legacy and recommends using the Chat Completions API for developing new applications.

See the following table for capabilities supported in Kong Gateway version v3.11+ or later:

Core text generation

The following reference tables detail feature availability across supported LLM providers when used with the AI Proxy plugin.

Support for chat, completions, and embeddings:

The following providers are supported by the legacy Completions API:

• OpenAI
• Azure OpenAI
• Cohere
• Llama2
• Amazon Bedrock
• Gemini
• Hugging Face

Advanced text generation v3.11+

Support for function calling, tool use, and batch processing:

Audio features v3.11+

Support for text-to-speech, transcription, and translation:

Image features v3.11+

Support for image generation, image editing interaction:

The AI Proxy plugin will mediate the following for you:

• Request and response formats appropriate for the configured config.targets[].model.provider and config.targets.route_type
• The following service request coordinates (unless the model is self-hosted):
  • Protocol
  • Host name
  • Port
  • Path
  • HTTP method
• Authentication on behalf of the Kong API consumer
• Decorating the request with parameters from the config.targets[].model.options block, appropriate for the chosen provider
• Recording of usage statistics of the configured LLM provider and model into your selected Kong log plugin output
• Optionally, additionally recording all post-transformation request and response messages from users, to and from the configured LLM
• Fulfillment of requests to self-hosted models, based on select supported format transformations

Flattening all of the provider formats allows you to standardize the manipulation of the data before and after transmission. It also allows your to provide a choice of LLMs to the Kong Gateway Consumers, using consistent request and response formats, regardless of the backend provider or model.

v3.11+ AI Proxy supports REST-based full-text responses, including RESTful endpoints such as llm/v1/responses, llm/v1/files, llm/v1/assisstants and llm/v1/batches. RESTful endpoints support CRUD operations— you can POST to create a response, GET to retrieve it, or DELETE to remove it.

The plugin’s route_type should be set based on the target upstream endpoint and model, based on this capability matrix:

The following requirements are enforced by upstream providers:

• For Azure Responses API, set config.azure_api_version to "preview".
• For OpenAI and Azure Assistant APIs, include the header OpenAI-Beta: assistants=v2.
• For requests with large payloads (e.g., image edits, audio transcription/translation), consider increasing config.max_request_body_size to three times the raw binary size.

The following upstream URL patterns are used:

While only the Llama2 and Mistral models are classed as self-hosted, the target URL can be overridden for any of the supported providers. For example, a self-hosted or otherwise OpenAI-compatible endpoint can be called by setting the same config.targets[].model.options.upstream_url plugin option.

v3.10+ If you are using each provider’s native SDK, Kong Gateway allows you to transparently proxy the request without any transformation and return the response unmodified. This can be done by setting config.llm_format to a value other than openai, such as gemini or bedrock. See the section below for more details.

In this mode, Kong Gateway will still provide useful analytics, logging, and cost calculation.

Input formats

Kong Gateway mediates the request and response format based on the selected config.targets[].model.provider and config.targets.route_type.

v3.10+ By default, Kong Gateway uses the OpenAI format, but you can customize this using config.llm_format. If llm_format is not set to openai, the plugin will not transform the request when sending it upstream and will leave it as-is.

The Kong Gateway AI Proxy accepts the following inputs formats, standardized across all providers. The config.targets.route_type must be configured respective to the required request and response format examples.

Text generation inputs

The following examples show standardized text-based request formats for each supported llm/v1/* route. These formats are normalized across providers to help simplify downstream parsing and integration.

{
    "messages": [
        {
            "role": "system",
            "content": "You are a scientist."
        },
        {
            "role": "user",
            "content": "What is the Boltzmann equation?"
        }
    ]
}

{
    "messages": [
        {
            "role": "system",
            "content": "You are a scientist."
        },
        {
            "role": "user",
            "content": "What is the Boltzmann equation?"
        }
    ],
    "guardrailConfig": {
        "guardrailIdentifier": "$GUARDRAIL-IDENTIFIER",
        "guardrailVersion": "1",
        "trace": "enabled"
    }
}

{
    "prompt": "You are a scientist. What is quantum entanglement?"
}

{
  "input": "Quantum computing is expected to revolutionize cryptography.",
  "model": "text-embedding-ada-002",
  "encoding_format": "float"
}

curl http://localhost:8000 \
  -F purpose="batch" \
  -F file="@mydata.jsonl"

{
  "instructions": "You are a frontend mentor. When asked a question, write and explain JavaScript code to help the user understand key concepts.",
  "name": "Frontend Mentor",
  "tools": [{"type": "code_interpreter"}],
  "model": "gpt-4o"
}

{
    "input_file_id": "file-abc123",
    "endpoint": "/v1/chat/completions",
    "completion_window": "24h"
}

{
  "input": "Summarize the key differences between HTTP/1.1 and HTTP/2."
}

Audio and image generation inputs

The following examples show standardized audio and image request formats for each supported route. These formats are normalized across providers to help simplify downstream parsing and integration.

curl http://localhost:8000 \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "input": "In the heart of the city, the rain whispered secrets to the streets.",
    "voice": "serene"
  }' \
  --output speech.mp3

curl http://localhost:8000/ \
  -H "Content-Type: multipart/form-data" \
  -F file="@/path/to/file/audio.mp3" \

curl http://localhost:8000/ \
  -H "Content-Type: multipart/form-data" \
  -F file="@/path/to/file/german.m4a" \

{
  "prompt": "An abstract visualization of API traffic flowing through a secure gateway",
  "n": 1,
  "size": "1024x512"
}

curl -s -D >(grep -i x-request-id >&2) \
  -o >(jq -r '.data[0].b64_json' | base64 --decode > breakfast-platter.png) \
  -X POST "https://api.openai.com/v1/images/edits" \
  -F "image[]=@pancakes.png" \
  -F "image[]=@coffee-cup.png" \
  -F "image[]=@fruit-bowl.png" \
  -F "image[]=@orange-juice.png" \
  -F 'prompt=Create a delicious breakfast platter with these four items arranged beautifully'

Response formats

Conversely, the response formats are also transformed to a standard format across all providers:

Text-based responses

{
    "choices": [
        {
            "finish_reason": "stop",
            "index": 0,
            "message": {
                "content": "The Boltzmann equation is a fundamental equation in statistical mechanics that describes how the distribution function f(x, v, t) of particles in a gas evolves over time. It accounts for the effects of particle collisions and external forces, providing a bridge between microscopic particle dynamics and macroscopic thermodynamic behavior. The general form of the equation is: ∂f/∂t + v · ∇f + F · ∇_v f = (∂f/∂t)_collision, where f is the distribution function, v is velocity, F is an external force, and the right-hand side represents the change in f due to collisions.",
                "role": "assistant"
            }
        }
    ],
    "created": 1707769597,
    "id": "chatcmpl-ID",
    "model": "gpt-4-0613",
    "object": "chat.completion",
    "usage": {
        "completion_tokens": 94,
        "prompt_tokens": 26,
        "total_tokens": 120
    }
}

{
    "choices": [
        {
            "finish_reason": "stop",
            "index": 0,
            "text": "Quantum entanglement is a phenomenon where particles become interconnected such that the state of one instantly influences the state of another, regardless of distance."
        }
    ],
    "created": 1707769597,
    "id": "cmpl-ID",
    "model": "gpt-3.5-turbo-instruct",
    "object": "text_completion",
    "usage": {
        "completion_tokens": 26,
        "prompt_tokens": 9,
        "total_tokens": 35
    }
}

{
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "embedding": [
        0.0023064255,
        -0.009327292,
        .... (1536 floats total for ada-002)
        -0.0028842222,
      ],
      "index": 0
    }
  ],
  "model": "text-embedding-ada-002",
  "usage": {
    "prompt_tokens": 8,
    "total_tokens": 8
  }
}

{
  "id": "file-abc123",
  "object": "file",
  "bytes": 120000,
  "created_at": 1677610602,
  "filename": "mydata.jsonl",
  "purpose": "fine-tune",
}

{
  "id": "batch_d41d8cd98f00b204e9800998ecf8427e",
  "object": "batch",
  "endpoint": "/v1/chat/completions",
  "errors": null,
  "input_file_id": "file-abcd",
  "completion_window": "24h",
  "status": "validating",
  "output_file_id": null,
  "error_file_id": null,
  "created_at": 1751281814,
  "in_progress_at": null,
  "expires_at": 1751368214,
  "finalizing_at": null,
  "completed_at": null,
  "failed_at": null,
  "expired_at": null,
  "cancelling_at": null,
  "cancelled_at": null,
  "request_counts": {
    "total": 0,
    "completed": 0,
    "failed": 0
  },
  "metadata": null
}

{
  "id": "asst_def456",
  "object": "assistant",
  "created_at": 1698984975,
  "name": "Frontend Mentor",
  "description": null,
  "model": "gpt-4o",
  "instructions": "You are a frontend mentor. When asked a question, write and explain JavaScript code to help the user understand key concepts.",
  "tools": [
    {
      "type": "code_interpreter"
    }
  ],
  "metadata": {},
  "top_p": 1.0,
  "temperature": 1.0,
  "response_format": "auto"
}

{
  "id": "resp_67ccd2bed1ec8190b14f964abc0542670bb6a6b452d3795b",
  "object": "response",
  "created_at": 1741476542,
  "status": "completed",
  "error": null,
  "incomplete_details": null,
  "instructions": null,
  "max_output_tokens": null,
  "model": "gpt-4.1-2025-04-14",
  "output": [
    {
      "type": "message",
      "id": "msg_67ccd2bf17f0819081ff3bb2cf6508e60bb6a6b452d3795b",
      "status": "completed",
      "role": "assistant",
      "content": [
        {
          "type": "output_text",
          "text": "HTTP/1.1 uses a single connection per request-response cycle, leading to inefficiencies, especially with multiple resources. In contrast, HTTP/2 supports multiplexing, allowing multiple streams over one connection, which reduces latency. HTTP/2 also introduces binary framing and header compression for improved performance.",
          "annotations": []
        }
      ]
    }
  ],
  "parallel_tool_calls": true,
  "previous_response_id": null,
  "reasoning": {
    "effort": null,
    "summary": null
  },
  "store": true,
  "temperature": 1.0,
  "text": {
    "format": {
      "type": "text"
    }
  },
  "tool_choice": "auto",
  "tools": [],
  "top_p": 1.0,
  "truncation": "disabled",
  "usage": {
    "input_tokens": 36,
    "input_tokens_details": {
      "cached_tokens": 0
    },
    "output_tokens": 60,
    "output_tokens_details": {
      "reasoning_tokens": 0
    },
    "total_tokens": 96
  },
  "user": null,
  "metadata": {}
}

Image, and audio responses

The following examples show standardized response formats returned by supported audio/ and image/ routes. These formats are normalized across providers to support consistent multimodal output parsing.

{
  "text": "Imagine the wildest idea that you've ever had, and you're curious about how it might scale to something that's a 100 or a 1,000 times bigger. This is a place where you can get to do that.",
  "usage": {
    "type": "tokens",
    "input_tokens": 14,
    "input_token_details": {
      "text_tokens": 0,
      "audio_tokens": 14
    },
    "output_tokens": 45,
    "total_tokens": 59
  }
}

{
  "text": "Hello, my name is Wolfgang and I come from Germany. Where are you heading today?"
}

{
  "created": 1713833628,
  "data": [
    {
      "b64_json": "..."
    }
  ],
  "usage": {
    "total_tokens": 100,
    "input_tokens": 50,
    "output_tokens": 50,
    "input_tokens_details": {
      "text_tokens": 10,
      "image_tokens": 40
    }
  }
}

{
  "created": 1713833629,
  "data": [
    {
      "b64_json": "..."
    }
  ],
  "usage": {
    "total_tokens": 100,
    "input_tokens": 50,
    "output_tokens": 50,
    "input_tokens_details": {
      "text_tokens": 10,
      "image_tokens": 40
    }
  }
}

The request and response formats are loosely modeled after OpenAI’s API. For detailed format specifications, see the sample OpenAPI specification.

When config.llm_format is set to a native format, only the corresponding provider is supported with its specific APIs as listed below.

Caveats and limitations

The following sections detail the provider and statistic logging limitations.

Provider-specific limitations

• Anthropic: Does not support llm/v1/completions or llm/v1/embeddings.
• Llama2: Raw format lacks support for llm/v1/embeddings.
• Bedrock and Gemini: Only support auth.allow_override = false.

Statistics logging limitations

• Anthropic: No statistics logging for llm/v1/completions.
• OpenAI and Azure: No statistics logging for assistants, batch, or audio APIs.
• Bedrock: No statistics logging for image generation or editing APIs.

The plugin allows you to substitute values in the config.targets[].model.name and any parameter under config.targets[].model.options with specific placeholders, similar to those in the Request Transformer Advanced plugin.

The following templated parameters are available:

• $(headers.header_name): The value of a specific request header.
• $(uri_captures.path_parameter_name): The value of a captured URI path parameter.
• $(query_params.query_parameter_name): The value of a query string parameter.

You can combine these parameters with an OpenAI-compatible SDK in multiple ways using the AI Proxy plugin, depending on your specific use case:

This can be used to OpenAI-compatible SDK with this plugin in multiple ways, depending on the required use case.