🐛 Patch Changes
- 3b345e7: Fix provider disconnect cleanup so removing an OAuth subscription clears routes pinned to that auth type even when an API-key credential for the same provider remains connected.
- 492cf46: Preserve provider key labels and priorities when duplicating agents.
- 045907d: Preserve native Google
generationConfigfields when routing requests to Gemini, while keeping existing OpenAI-compatible generation aliases as explicit overrides. - 9438035: Serve model parameter specs from the MPS catalog API and support scoped per-route defaults.
- 4e780be: Make saved Manifest model parameters authoritative over overlapping client request parameters while preserving client parameters that Manifest does not configure.
- bea267e: Fetch model parameter specs per-model on demand instead of downloading the full MPS catalog on every Routing-page load, and add an ETag conditional GET to the catalog refresh. Keeps the dashboard payload flat as the catalog grows.
- 680feca: Stop slow startup model-catalog syncs from stalling boot past the deploy healthcheck. The OpenRouter, models.dev, GitHub, and modelparameters.dev fetches now run in the background instead of blocking
app.listen(), and the two that lacked a timeout (OpenRouter, GitHub) now abort after 10s. The pricing cache still warms up with real data once those fetches land.