llm_gateway.yaml

Source: config/llm_gateway.yaml

# LLM Gateway Configuration — editable via dashboard, hot-reloaded on save.

scheduling:
  active_hours: "06:00-22:00"
  schedule_drain_minutes: 2

queue:
  max_external_depth: 20
  max_interrupt_count: 3

priority_map:
  parse_task: critical
  challenge_task: critical
  generate_digest: normal
  extract_preferences: normal
  dedup_check: normal
  prep_research: normal
  task_decompose: normal
  generate_nudge: background
  generate_reminder: background
  generate_weekly_digest: normal

rate_limits:
  default:
    requests_per_minute: 10
    requests_per_hour: 100
  callers: {}

budget:
  daily_external_usd: 5.00
  alert_pct: 80

cloud:
  max_per_request_usd: 0.50
  daily_cloud_external_usd: 2.00

alerts:
  queue_depth_warning: 10
  rate_limit_alert_threshold: 3
  debounce_minutes: 10

ollama_health_check: true

gpu:
  home_model: "qwen2.5:32b-instruct-q6_K"
  swap_timeout_s: 120
  restore_home_delay_s: 30
  alerts:
    swaps_per_hour_warning: 4
    swap_wait_ms_warning: 60000
    swap_overhead_pct_warning: 25