Chat Completions To Gemini

# Chat Completions to Vertex AI Gemini
#
# Transforms OpenAI Chat Completions requests into Vertex AI Gemini
# generateContent format and translates responses back.
#
# Request: messages flattened to Gemini contents, system messages
# hoisted to systemInstruction, tools mapped to function
# declarations, generationConfig built from OpenAI parameters.
#
# Response: candidates mapped to choices, usageMetadata to usage,
# finishReason normalized. Streaming SSE frames are translated
# inline from Gemini streamGenerateContent to Chat Completions
# chunk format.
#
# The filter rewrites the upstream path to the Vertex AI
# generateContent (or streamGenerateContent) endpoint using the
# model name from the request body and the configured project and
# region.
#
# Production deployments should pair this filter with `gcp_adc`
# (requires --features gcp-adc-filter) for automatic OAuth2 token
# injection and TLS to the regional Vertex AI endpoint
# (e.g. us-central1-aiplatform.googleapis.com:443).

listeners:
  - name: vertex-gateway
    address: "127.0.0.1:8080"
    filter_chains: [vertex-gemini]

filter_chains:
  - name: vertex-gemini
    filters:
      - filter: openai_chat_completions_to_vertexai_gemini
        project: my-gcp-project
        region: us-central1
        # max_body_bytes: 1048576  # default: 1 MiB

      - filter: router
        routes:
          - path_prefix: "/"
            cluster: vertex-backend

      - filter: load_balancer
        clusters:
          - name: vertex-backend
            endpoints:
              - "127.0.0.1:3000"

insecure_options:
  allow_private_endpoints: true # example proxies to a local backend