* Mark a model as temporarily overloaded (529/503 capacity). * Shorter cooldown than rate limits since capacity restores quickly.
(modelId: string)
| 519 | * Shorter cooldown than rate limits since capacity restores quickly. |
| 520 | */ |
| 521 | function markOverloaded(modelId: string): void { |
| 522 | overloadedModels.set(modelId, Date.now()); |
| 523 | console.log(`[ClawRouter] Model ${modelId} overloaded, will deprioritize for 15s`); |
| 524 | } |
| 525 | |
| 526 | /** Check if a model is in its overload cooldown period. */ |
| 527 | function isOverloaded(modelId: string): boolean { |