You can now set a global default template for AI enhancements via general.selected_template_id, and new summaries will automatically use it when you don’t explicitly pick a template. Regenerate/retry flows also preserve the original note’s template_id, so you won’t get surprise template switches. 🚀

Cactus STT models are now a functional beta—replacing the “Coming soon” UI with real download/delete/select flows plus a new Cloud Handoff toggle. Under the hood, server start/stop is centralized in registry.ts to prevent redundant restarts when switching models/providers, and Tauri commands/permissions were added to fix things like the “open folder” action. This makes local Cactus transcription feel much more real-world usable (and less flaky) for users testing the beta.

Cloud AI settings got a cleanup: duplicated STT/LLM upgrade CTA logic is now a shared HyprCloudCTAButton, with more consistent “Pro (Cloud)” labeling. Non‑Pro users are now gated when trying to pick the hyprnote LLM provider (selection triggers the upgrade flow instead of silently switching), and the hyprnote accordion auto-opens when no provider is configured for a smoother onboarding UX.

Swapped out the homegrown PostHog + feature flag crates for the official posthog-rs 0.4.1 client, cutting a bunch of custom code and simplifying the stack. Users/devs get a richer flag API (multivariate + payloads) and optional local flag evaluation with caching for faster, fewer-network lookups—now shared via the analytics plugin instead of duplicated state.

Fixes a bug where cloud LLMs sometimes prepend text like “Here is the summary:” before the first heading, causing the enhance stream to be rejected. The validator pipeline now strips preamble text before running H1/section heading checks, making generated summaries more reliable for users.