server:
  host: "127.0.0.1"
  port: 18080
  admin_port: 18081
  max_request_body_mb: 5

auth:
  enabled: true
  methods: [api_key]

scheduler:
  max_running_tasks: 2
  max_queued_tasks: 10
  fairness: weighted_fair_queue
  priority_aging_seconds: 5
  reserved_realtime_slots: 1

timeouts:
  default_connect_ms: 2000
  default_queue_ms: 2000
  default_first_token_ms: 5000
  default_inference_ms: 10000
  default_idle_ms: 5000
  default_total_ms: 15000
  cancel_grace_period_ms: 1000

context:
  safety_margin_ratio: 0.1
  default_policy: recent_only
  max_session_messages: 20
  session_idle_ttl_minutes: 5
  enable_prompt_persistence: false

models:
  test-chat:
    provider: ollama
    actual_model: qwen2.5:0.5b
    endpoint: http://127.0.0.1:11434
    context_window: 4096
    max_output_tokens: 256
    max_concurrency: 1
    residency: always
    cancel_supported: true

routing:
  sensitive_data_local_only: true
  allow_cloud_fallback_by_default: false
  overload_strategy:
    - smaller_local_model
    - reject

circuit_breaker:
  error_rate_threshold: 0.2
  min_requests: 3
  window_seconds: 30
  open_duration_seconds: 10
  half_open_max_requests: 1

backpressure:
  level1_threshold: 0.70
  level2_threshold: 0.85
  level3_threshold: 0.95

observability:
  metrics_enabled: true
  metrics_path: /metrics
  tracing_enabled: false
  prompt_logging: metadata_only
  audit_retention_days: 7
  log_level: debug

storage:
  session_db: sqlite:///tmp/edgeai-test/sessions.db
  task_state: sqlite:///tmp/edgeai-test/tasks.db
  redis:
    enabled: false