🐛 Patch Changes
-
35b326c: Add migration to backfill tenant_id on agent_messages from tenants table
-
e4f5b10: Drop 4 unused composite indexes on write-only OTLP ingestion tables (tool_executions, token_usage_snapshots, cost_snapshots, agent_logs) to reduce write amplification
-
45dd667: Fix missing canonical provider models in local mode and improve routing page UX
- Create canonical model entries (e.g.
gemini-2.5-prowith provider "Google") during pricing sync when no seeded entry exists, fixing local mode showing only OpenRouter-branded models - Replace full tier list refetch with local state mutations on model change, reset, and fallback operations for instant UI updates
- Add loading indicator ("Changing...") on the Change button while model override is saving
- Add toast notification when removing a fallback model
- Create canonical model entries (e.g.
-
3143303: Fix token values showing as 0 for OTLP-ingested messages by capturing usage directly from proxy responses
-
8ec8b25: Fix routing model picker showing non-curated models from OpenRouter
-
78e8ec2: Progressive rendering for Routing page skeleton loading state
-
a1dd405: Speed up Overview and Messages pages
- Parallelize messages endpoint DB queries (count, data, models run concurrently via Promise.all)
- Show stale data during SSE refetches instead of flashing skeleton loaders
- Debounce cost filter inputs on Messages page (400ms) to prevent rapid-fire API calls