Archivarix Tube Search

Hey IH — just launched Archivarix Tube Search after working on it on and off for ~2 years. It's a search engine for archived YouTube content — including videos that have been deleted, removed, region-blocked, or DMCA'd. We index ~1.5 billion videos from 2005 onwards by aggregating Internet Archive's Wayback Machine, Common Crawl. For any video ID you can pull metadata, thumbnails, and original captions where the archive captured them. We layered AI-generated summaries (Gemma 4 27B on a self-hosted vLLM box) and full audio transcription (Qwen3-ASR) for archived audio. There's also an MCP server (`npx tube-search-mcp`) so AI agents like Claude and Cursor can use search as a tool. Stack: Fastify 5 + Postgres + React + Tailwind 4. Self-hosted GPU box for the LLM/ASR stuff. Hetzner + 20 TB Storage Box for media. Free tier without signup for basic search.