1. Free first, always.
Priority order: GLM Flash (∞) → Gemini Flash (1,500) → Gemini Flash-Lite (1,000) → Groq (1,000) → OpenRouter free (50) → OpenCode free (bonus).
Paid lanes (DeepSeek/Kimi) only fire when every free lane is exhausted or limited.
2. Quota done → hide from customers.
When a lane hits 0 remaining, it flips to LIMITED and is removed from routing. Customers never see the error — the router silently uses the next lane.
3. Reset timer.
Each lane has its own reset schedule:
• Gemini: 00:00 Pacific (07:00 UTC)
• Groq / OpenRouter: 00:00 UTC
• OpenCode free: rolling 24h
• GLM: no cap (never resets needed).
When the timer fires, the lane auto-returns to ACTIVE.
4. Admin visibility.
This dashboard shows live status, remaining quota, countdown, and key health. Data source: free_quota_manager.py state file. Expose via the router's admin endpoint on the VPS.