# Token Rate Limiting -- Soft Over-Quota Enforcement (ai#1241)
#
# Per-rule `enforcement` chooses what happens when the algorithm denies a
# reservation because the token budget is exhausted:
#
#   hard   (default)  → 429 + X-RateLimit-*-Tokens response headers
#   soft              → forward the request and annotate it for downstream
#                       handling (deprioritize, degrade, meter-only, etc.)
#
# Soft requires an `over_quota:` annotation surface. Static headers are
# always set; optional `include_remaining` / `include_used` add dynamic
# quota metadata from the backend denial snapshot (`remaining`, and
# `limit - remaining` for used).
#
# Soft over-quota forwards are not reserved or reconciled: the window is
# not charged for that traffic (annotate only). Protective denials
# (invalid key, key/reservation capacity) still 429 even under soft.
#
# Graduated S1 tiers (see token-rate-limit-soft-tiers.yaml) still annotate
# *admitted* traffic below capacity. Soft here covers the path *above*
# capacity that would otherwise be a hard 429.
#
# Usage:
#   cargo run -p praxis-ai-proxy --features token-rate-limit-filter -- \
#     -c examples/configs/token-rate-limit-soft-enforcement.yaml
#
# Supported token_count provider values:
#   openai | anthropic | google | bedrock | azure

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

filter_chains:
  - name: main
    filters:
      - filter: router
        routes:
          - path: "/v1/chat/completions"
            cluster: backend

      - filter: token_rate_limit
        rules:
          # Soft: over-quota requests continue with annotation.
          - name: team-soft
            match:
              headers:
                x-app-id: soft
            algorithm: sliding_window
            window: 1h
            capacity: 100000
            reserved_tokens: 500
            enforcement: soft
            over_quota:
              headers:
                X-Over-Quota: "true"
                X-Token-Priority: low
              include_remaining: true
              include_used: true

          # Hard (default): exhausted budget → 429.
          - name: team-hard
            match:
              headers:
                x-app-id: hard
            algorithm: sliding_window
            window: 1h
            capacity: 50000
            reserved_tokens: 200

      - filter: token_count
        provider: openai

      - filter: access_log

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

insecure_options:
  allow_private_endpoints: true