Azure API Management AI Gateway Chat and Embed

arazzo: 1.0.1
info:
  title: Azure API Management AI Gateway Chat and Embed
  summary: Send a chat completion through the AI gateway, then embed the same prompt for retrieval.
  description: >-
    A retrieval-augmented pattern routed entirely through the API Management AI
    gateway. The workflow sends a chat completion request to a deployment to get
    an assistant answer, then sends the original user prompt to an embeddings
    deployment so it can be stored in a vector index for later semantic search.
    Every step spells out its request inline so the flow can be read and
    executed without opening the underlying OpenAPI description.
  version: 1.0.0
sourceDescriptions:
- name: aiGatewayApi
  url: ../openapi/microsoft-azure-api-management-ai-gateway-openapi.yaml
  type: openapi
workflows:
- workflowId: ai-gateway-chat-and-embed
  summary: Get a chat completion and an embedding through the AI gateway.
  description: >-
    Sends a chat completion request to a chat deployment, then sends the prompt
    to an embeddings deployment to produce a vector.
  inputs:
    type: object
    required:
    - chatDeploymentId
    - embeddingDeploymentId
    - prompt
    properties:
      chatDeploymentId:
        type: string
        description: The chat deployment identifier (e.g. gpt-4o-deployment).
      embeddingDeploymentId:
        type: string
        description: The embeddings deployment identifier.
      prompt:
        type: string
        description: The user prompt to answer and embed.
      maxTokens:
        type: integer
        description: The maximum number of completion tokens.
  steps:
  - stepId: chatCompletion
    description: >-
      Send the prompt as a chat completion request to the chat deployment and
      capture the assistant answer.
    operationId: AIGateway_ChatCompletions
    parameters:
    - name: deployment-id
      in: path
      value: $inputs.chatDeploymentId
    requestBody:
      contentType: application/json
      payload:
        messages:
        - role: system
          content: You are a helpful assistant.
        - role: user
          content: $inputs.prompt
        max_tokens: $inputs.maxTokens
        temperature: 0.7
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      answer: $response.body#/choices/0/message/content
      totalTokens: $response.body#/usage/total_tokens
  - stepId: embedPrompt
    description: >-
      Send the original prompt to the embeddings deployment to produce a vector
      for indexing and semantic search.
    operationId: AIGateway_Embeddings
    parameters:
    - name: deployment-id
      in: path
      value: $inputs.embeddingDeploymentId
    requestBody:
      contentType: application/json
      payload:
        input: $inputs.prompt
        model: text-embedding-ada-002
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      embedding: $response.body#/data/0/embedding
  outputs:
    answer: $steps.chatCompletion.outputs.answer
    embedding: $steps.embedPrompt.outputs.embedding