v1.0.0

vimmary v1.0.0

YouTube video summary service. Receives bookmarks via webhooks or manual URL submission, fetches transcripts, generates LLM summaries, and stores everything in Postgres + pgvector for semantic search.

Highlights

• Automatic video processing — Karakeep webhooks trigger transcript fetch → LLM summary → storage
• Manual URL submission — paste any YouTube URL in the web UI
• Two summary providers — Claude API or Mistral, configurable per deployment
• Two detail levels — medium (default) and deep (on-demand)
• Hybrid search — keyword + semantic search with Reciprocal Rank Fusion (RRF)
• MCP server — 6 tools for searching, browsing, and managing summaries from Claude Desktop or other MCP clients
• Web UI — React frontend embedded in the Go binary (Videos, Stats, Settings pages)
• Multi-user support — per-user video libraries with Tailscale authentication
• Karakeep integration — bidirectional sync (webhook ingest + summary writeback), bulk import, backlinks
• Processing queue — adaptive rate limiting, auto-retry with exponential backoff, batch retry
• Docker image — meltforce/vimmary:latest / meltforce/vimmary:v1.0.0

Getting started

See the README for a complete quick start guide with Docker Compose.

Requirements

• PostgreSQL 16 with pgvector (pgvector/pgvector:pg16)
• Mistral API key (embeddings)
• Claude or Mistral API key (summaries)
• Optional: Tailscale for authentication, Karakeep for bookmark integration