chat:
  default_model: production
  backend:
    type: litellm_router
  routing:
    engine: litellm_router
    strategy: weighted
    circuit_breaker:
      enabled: true
      failure_threshold: 4
      recovery_timeout_seconds: 30
  cache:
    enabled: false
  observability:
    enabled: true
    privacy:
      log_inputs: false
      log_outputs: false
  security:
    allowed_providers: [azure_openai, anthropic]
    allowed_base_url_hosts: [example.openai.azure.com]
  models:
    production:
      aliases: [chat-default]
      fallbacks: [disaster-recovery]
      deployments:
        - name: azure-east
          provider: azure_openai
          model: azure/REPLACE_WITH_AZURE_DEPLOYMENT
          deployment_name: REPLACE_WITH_AZURE_DEPLOYMENT
          api_key: ${AZURE_OPENAI_API_KEY}
          api_base: ${AZURE_OPENAI_ENDPOINT}
          api_version: ${AZURE_OPENAI_API_VERSION}
          weight: 3
          capabilities: &chat_capabilities
            streaming: true
            structured_output: true
            json_mode: true
            tools: true
        - name: azure-west
          provider: azure_openai
          model: azure/REPLACE_WITH_AZURE_DEPLOYMENT
          deployment_name: REPLACE_WITH_AZURE_DEPLOYMENT
          api_key: ${AZURE_OPENAI_API_KEY}
          api_base: ${AZURE_OPENAI_ENDPOINT}
          api_version: ${AZURE_OPENAI_API_VERSION}
          weight: 1
          capabilities: *chat_capabilities
    disaster-recovery:
      deployments:
        - name: anthropic-dr
          provider: anthropic
          model: anthropic/REPLACE_WITH_CHAT_MODEL
          api_key: ${ANTHROPIC_API_KEY}
          capabilities: *chat_capabilities

embed:
  default_model: production
  routing:
    engine: harbor
    strategy: least_busy
  security:
    allowed_providers: [azure_openai]
    allowed_base_url_hosts: [example.openai.azure.com]
  models:
    production:
      embedding_space: harbor-index-v1
      deployments:
        - name: azure-embedding-east
          provider: azure_openai
          model: azure/REPLACE_WITH_EMBEDDING_DEPLOYMENT
          deployment_name: REPLACE_WITH_EMBEDDING_DEPLOYMENT
          api_key: ${AZURE_OPENAI_API_KEY}
          api_base: ${AZURE_OPENAI_ENDPOINT}
          api_version: ${AZURE_OPENAI_API_VERSION}
          expected_dimensions: 1536
          capabilities: &embed_capabilities
            batch: true
            configurable_dimensions: true
            default_dimensions: 1536
            encoding_format: true
        - name: azure-embedding-west
          provider: azure_openai
          model: azure/REPLACE_WITH_EMBEDDING_DEPLOYMENT
          deployment_name: REPLACE_WITH_EMBEDDING_DEPLOYMENT
          api_key: ${AZURE_OPENAI_API_KEY}
          api_base: ${AZURE_OPENAI_ENDPOINT}
          api_version: ${AZURE_OPENAI_API_VERSION}
          expected_dimensions: 1536
          capabilities: *embed_capabilities

rerank:
  default_model: production
  routing:
    engine: harbor
    strategy: ordered
  security:
    allowed_providers: [bedrock, cohere]
  models:
    production:
      fallbacks: [disaster-recovery]
      default_params: &rerank_defaults
        top_n: 10
        return_documents: false
      deployments:
        - name: bedrock-primary
          provider: bedrock
          model: bedrock/REPLACE_WITH_RERANK_MODEL
          aws_region_name: ${AWS_REGION}
          allow_ambient_credentials: true
    disaster-recovery:
      default_params: *rerank_defaults
      deployments:
        - name: cohere-dr
          provider: cohere
          model: cohere/REPLACE_WITH_RERANK_MODEL
          api_key: ${COHERE_API_KEY}

profiles:
  development:
    chat:
      timeouts:
        request_seconds: 30
      observability:
        enabled: false
  production:
    chat:
      timeouts:
        request_seconds: 90
    embed:
      timeouts:
        request_seconds: 90
    rerank:
      timeouts:
        request_seconds: 90