# External Metering
#
# Pre-request balance check and post-response token usage reporting
# against an external metering service. Identity headers are captured
# from configurable tenant headers (default prefix: x-tenant-) and
# stripped before forwarding upstream.
#
# Usage:
#   cargo run -p praxis-ai-proxy -- -c examples/configs/external-metering.yaml
#   curl -X POST http://localhost:8080/v1/chat/completions \
#     -H "Content-Type: application/json" \
#     -H "x-tenant-username: alice" \
#     -H "x-tenant-group: engineering" \
#     -H "x-tenant-subscription: sub-42" \
#     -d '{"model":"gpt-4","messages":[{"role":"user","content":"hi"}]}'
#
# Pipeline ordering matters: external_metering is declared before
# token_count so that response hooks (which run in reverse order)
# execute token_count first, writing token.input / token.output /
# token.total to filter_metadata, then external_metering reads them
# and sends a CloudEvent to the metering service.
#
# Balance check endpoint:
#   GET {metering_url}/api/v1/customers/{username}/entitlements/{feature_key}/value?model={model}
#
# Usage report endpoint:
#   POST {metering_url}/api/v1/events  (CloudEvents 1.0 JSON)

listeners:
  - name: default
    address: "127.0.0.1:8080"
    filter_chains:
      - main

filter_chains:
  - name: main
    filters:
      - filter: router
        routes:
          - path_prefix: "/"
            cluster: backend

      - filter: external_metering
        metering_url: "http://127.0.0.1:9090"
        # The metering service in this example listens on loopback, a
        # non-public address, so callouts must be explicitly opted in.
        allow_private_endpoint: true
        timeout_seconds: 5
        feature_key: "inference-tokens"
        source: "ai-gateway"
        fail_open: true
        identity_header_prefix: "x-tenant-"
        # Namespace the identity_header_guard filter writes captured headers
        # under; must match that filter's metadata_namespace when both run.
        # identity_metadata_namespace: "identity"
        # Optional fallbacks for deployments where an upstream authentication
        # layer does not inject identity headers. Without default_username,
        # requests carrying no identity header are not metered at all.
        # default_username: "anonymous"
        # default_model: "unknown"

      - filter: token_count
        provider: openai

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

insecure_options:
  allow_private_endpoints: true # example proxies to local backends