2026-03-25: Shared AI Client and Voice-to-Memory Port

What Changed

• added a shared AI client at src/server/ai-client.ts
• refactored Flashcard Maker onto that shared AI client
• ported Voice-to-Memory from the active frain line into codex/local-dev
• added voice_recordings persistence
• added a reusable src/server/plugins/ports/voice-to-memory.ts module
• added /api/v1/plugins/voice-to-memory
• added /app/voice plus navigation and command-palette entry points
• added unit coverage for AI config resolution and voice-title generation

Why It Matters

This closes two important convergence gaps at once:

• AI-heavy plugins now have a shared provider-resolution layer
• audio/transcription features now have a codex-native reference port

That makes future ports easier because the next AI plugins can reuse both:

• the shared AI client
• the Voice-to-Memory media workflow pattern

Remaining Gaps

• several existing AI-heavy routes still need migration onto the shared AI client
• Voice-to-Memory stores transcript-first metadata today rather than retaining raw audio assets
• richer playback, uploads beyond browser recording, and deeper background processing are still follow-up work