What's New
Rate Limit Handling Overhaul
Comprehensive improvements to handle high-concurrency scenarios and prevent "Max retries exceeded" errors:
- Emergency fallback tiers: When all accounts become unhealthy, the proxy now uses "least bad" accounts instead of failing
- Short rate limit handling: Sub-second rate limits (< 1s) are now waited out instead of triggering account switches
- Progressive backoff: Model capacity exhaustion now uses tiered backoff (5s → 10s → 20s → 30s → 60s)
- Faster health recovery: Accounts recover from rate limit penalties 5x faster (10 points/hour vs 2)
- Throttled fallback requests: Emergency/lastResort modes add small delays to prevent API overload
Manual OAuth Authorization Mode
Added support for manually entering OAuth authorization codes in the WebUI, enabling setup on headless servers or environments where browser redirects don't work.
Contributors
Thanks to @mintfog for adding manual OAuth authorization mode!