Self-host Anagnorisis

Anagnorisis is a local-first recommendation platform that runs entirely on your own machine or home server. You rate files (images, music, text, videos) on a 0–10 scale, and the system trains a personal evaluator model from those ratings to predict scores for unrated content. Search works in three modes: filename fuzzy matching, content-based semantic search via embeddings, and metadata-based search using EXIF/ID3 tags and optional .meta sidecar files that enable distributed semantic indexing without a central protocol. The core engine (anagnorisis_core) is a standalone Python package with a CLI (anag) so the same models and memory can be shared between the web UI, command line, and a lightweight data server for file sharing. Built-in modules cover Images, Music, Text, Videos, and a Train UI; experimental external modules add web and YouTube indexing. Deployment is via Docker Compose with NVIDIA GPU support (CUDA), AMD ROCm, or CPU-only fallback. The project recommends a recent NVIDIA GPU with at least 8 GB VRAM and 32 GB RAM for best performance, plus ~20–25 GB disk space for models and caches. All processing stays local — no data leaves your machine unless you explicitly share .meta files.