Skip to content

v2026.04.5a - Got Local?

Choose a tag to compare

View all tags
@tophers tophers released this 23 Apr 00:06
· 62 commits to main since this release
v2026.04.5a
748ffa6

MixerBee Release v2026.04.5 - The "Got Local?" Update

revision a: bugfixes

This release focuses on privacy, and performance. Introduced support for local LLMs via Ollama (Read: I got annoyed with Gemini Limits), overhauled the vector indexing for larger libraries, and modernized the backend architecture.

New Features

  • Local AI Support (Ollama): You can now swap Google Gemini for a local Ollama instance and power the "Smart Builder" with local models like Qwen or Mistral without your library metadata ever leaving your network. Most testing done with Qwen2.5:7B and Qwen3:14B.
  • "Vibe" Blocks: A new UI block type. When the AI generates a selection based on a "mood" or "feeling," MixerBee creates a specialized Vibe Block. These blocks are AI-curated and display semantic matches found directly in your library.
  • Smart Settings Management: Settings are no longer just file-based. Added a database-backed settings system that syncs with your .env file automatically.
  • Delta Vector Indexing: The ChromaDB is now much smarter. Instead of re-indexing your entire library, it now performs a differential sync—adding only new items and removing deleted ones from the ChromaDB vector store.
  • Provider Badge: The Prompt area now clearly displays your active provider (Gemini or Ollama) and the specific model being used, if used, still completely optional.

Technical Improvements

  • FastAPI Modernization: Replaced legacy @app.on_event handlers with the modern lifespan context manager for cleaner startup/shutdown logic.
  • Enhanced Pydantic Models: Transitioned to Pydantic V2 patterns, utilizing AliasChoices for more resilient AI-to-Frontend data mapping.
  • Paginated Aggregation: Metadata collection for studios and genres now uses a paginated chunking strategy, preventing timeouts on extremely large Emby/Jellyfin libraries.
  • Centralized Logging: Introduced a dedicated logger.py utility with a VERBOSE_LOGGING toggle in the settings menu for easier troubleshooting. Add VERBOSE_LOGGING="1" (or True) to your .env to get a lot more logging out of MixerBee, helpful if you want to know what the AI is doing and why. (I will eventualy add this to the UI settings modal).

Refinements & Fixes

  • Footer UI Fix: The Action Bar is now more stable with improved CSS layering (has-fixed-footer) to prevent layout shift during block additions.
  • ID Coercion Fix: The backend now includes a regex-based ID cleaner to ensure that AI-generated IDs (sometimes wrapped in brackets or quotes) are correctly sanitized before being sent to the media server.
  • Scheduler Fix Added a $nextTick fix to the scheduler UI, ensuring the edit modal opens instantly when creating a new automated job.
  • Docker Support Improvements: Switched python build over to Trixie to have build-essentials availabile. As well as a better testing pipelines as I noticed Docker builds are often broken. Also added an ARM64 build to the workflow for Apple Silicon mostly.
Assets 2
Loading