request-validate.yaml

# Responses API Request Validation
#
# Validates Responses API requests and rejects invalid parameter
# combinations.
#
# Requires the ai-inference feature:
#   cargo build -p praxis-proxy --features ai-inference
#
# The openai_responses_format classifier must run before openai_responses_validate.
# It buffers the body, classifies the request as Responses API, and
# promotes model/stream/store/background to openai_responses_format.*
# metadata that openai_responses_validate reads for parameter-combination
# validation.
#
# openai_responses_validate then extracts conversation.id when present,
# generates response and conversation IDs, and writes responses.*
# metadata for downstream pipeline filters.
#
# Valid requests are forwarded to the upstream backend with the
# original body unchanged. Invalid requests receive a 400 JSON
# error response.
#
# Example valid request:
#
#   curl -X POST http://localhost:8080/v1/responses \
#     -H "Content-Type: application/json" \
#     -d '{"model":"gpt-4.1","input":"Hello, world!"}'
#
# Example rejected request (stream + background conflict):
#
#   curl -X POST http://localhost:8080/v1/responses \
#     -H "Content-Type: application/json" \
#     -d '{"model":"gpt-4.1","input":"test","stream":true,"background":true}'

listeners:
  - name: responses-gateway
    address: "127.0.0.1:8080"
    filter_chains:
      - validate-and-route

filter_chains:
  - name: validate-and-route
    filters:
      - filter: openai_responses_format
        on_invalid: reject
      - filter: openai_responses_validate
      - filter: router
        routes:
          - path: "/v1/responses"
            cluster: inference
      - filter: load_balancer
        clusters:
          - name: inference
            endpoints:
              - "127.0.0.1:8000"