{
  // pi-offload-router settings
  //
  // enabled:
  //   true enables offloading; false disables offload behavior but keeps /offload commands available.
  //
  // defaults.model:
  //   Shared fallback model for any offload whose model is set to "default".
  //   Allowed values:
  //   - "main" = use Pi's current main session model
  //   - "provider/modelId" = use a specific model directly
  //
  // defaults.taskTimeoutSeconds:
  //   Global maximum time for an offload model request after it starts running.
  //
  // defaults.queueTimeoutSeconds:
  //   Global maximum time an offload waits in the queue before it gets a concurrency slot.
  //
  // defaults.maxTokens:
  //   Global maximum output tokens the offload model may generate.
  //
  // offloads.<name>.model:
  //   Per-offload model override.
  //   Allowed values:
  //   - "default" = use defaults.model
  //   - "main" = use Pi's current main session model
  //   - "provider/modelId" = use a specific model directly
  //
  // offloads.compaction.taskTimeoutSeconds / queueTimeoutSeconds / maxTokens:
  //   Optional compaction-specific overrides for the global defaults.
  //
  // offloads.branchSummary / offloads.titleGeneration / offloads.handoff:
  //   These currently inherit defaults.taskTimeoutSeconds, defaults.queueTimeoutSeconds,
  //   and defaults.maxTokens unless you explicitly add overrides.
  //
  // concurrency.maxInFlight:
  //   Maximum number of offload requests running at once.
  "enabled": true,
  "defaults": {
    "model": "lmstudio/LOADED",
    "taskTimeoutSeconds": 300,
    "queueTimeoutSeconds": 600,
    "maxTokens": 8192
  },
  "offloads": {
    "compaction": {
      "model": "default",
      "taskTimeoutSeconds": 600,
      "queueTimeoutSeconds": 600,
      "maxTokens": 16384
    },
    "branchSummary": {
      "model": "default"
    },
    "titleGeneration": {
      "model": "default"
    },
    "handoff": {
      "model": "default"
    }
  },
  "concurrency": {
    "maxInFlight": 1
  }
}