Tool Routing

# Tool Composition Routing
#
# Demonstrates using `openai_tool_parse` to route Responses API requests
# by their tool composition. All requests go to the same inference
# backend — the branch conditions select different processing
# chains based on what tools are present in the request body.
#
# The `openai_tool_parse` filter parses the `tools` array and `tool_choice`
# from Responses API create requests and promotes summary facts to
# metadata and filter results. These promoted facts drive branch
# conditions and downstream filter decisions.
#
# In this example:
#
#   - Requests with `web_search` tools branch to a chain that could
#     apply search-specific policies (rate limiting, cost tracking,
#     extended timeouts) before forwarding to inference.
#
#   - All other requests (with or without tools) fall through to the
#     default chain.
#
# Example requests:
#
#   # Function tool — default chain, has_tools=true header set
#   curl -X POST http://localhost:8080/v1/responses \
#     -H "Content-Type: application/json" \
#     -d '{"model":"gpt-4.1","input":"What is 2+2?","tools":[{"type":"function","name":"calc"}]}'
#
#   # Web search tool — branches to web-search chain
#   curl -X POST http://localhost:8080/v1/responses \
#     -H "Content-Type: application/json" \
#     -d '{"model":"gpt-4.1","input":"Latest news","tools":[{"type":"web_search"}]}'
#
#   # No tools — default chain, no tool headers set
#   curl -X POST http://localhost:8080/v1/responses \
#     -H "Content-Type: application/json" \
#     -d '{"model":"gpt-4.1","input":"Hello, world!"}'

listeners:
  - name: ai-gateway
    address: "127.0.0.1:8080"
    filter_chains: [tool-routing]

filter_chains:
  - name: tool-routing
    filters:
      - filter: openai_tool_parse
        branch_chains:
          - name: web-search
            on_result:
              filter: openai_tool_parse
              key: has_web_search
              result: "true"
            chains:
              - name: web-search-chain
                filters:
                  # Placeholder: insert search-specific filters here
                  # (rate limiting, cost tracking, extended timeouts).
                  - filter: router
                    routes:
                      - path_prefix: "/"
                        cluster: "inference"
            rejoin: shared_load_balancer

      # Default: all other requests (with or without tools)
      - filter: router
        routes:
          - path_prefix: "/"
            cluster: "inference"
      - name: shared_load_balancer
        filter: load_balancer
        clusters:
          - name: "inference"
            endpoints:
              - "127.0.0.1:3001"

insecure_options:
  allow_private_endpoints: true # example proxies to local backends