Anagnorisis
A self-hosted recommendation platform that builds a personalized content-discovery model on your own machine, trained only on your own ratings.
🔗 Visit Anagnorisis
Description
Recommendation algorithms on YouTube, Spotify, or TikTok decide what you see next based on what keeps you engaged — not necessarily what you'd actually rate highly. Anagnorisis is a self-hosted alternative: it builds a personalized recommendation model that runs entirely on your own machine, trained only on how you rate your own content, with no platform pulling the strings.
It supports three search modes — filename fuzzy matching, content-based embeddings, and semantic metadata search — plus a 0-10 rating system that trains your personal preference model over time. A 'soft hash' identification system keeps ratings attached to files even after you move or rename them, and distributed indexing via metadata sidecar files lets you search a remote library without downloading everything first. It handles images, music, text, and video through a modular architecture, with both a CLI (anag) and a web interface, and runs via Docker. It's open-source under AGPL-3.0, with 341 GitHub stars and recent, active commits.
💬 Our review
The short version: if you're uncomfortable with how much control commercial recommendation algorithms have over what you watch or listen to, and you're willing to self-host, Anagnorisis gives you a genuinely local, transparent alternative — for free.
There's no direct commercial equivalent, since the entire pitch is running the recommendation model yourself instead of trusting a platform's black-box algorithm — that's a meaningful tradeoff against YouTube/Spotify's convenience and scale. The hardware requirements are real: a minimum 8GB VRAM GPU, 32GB RAM, and 20-25GB of disk space put it out of reach of casual users, and there's a learning curve to set up and train it well. For a media hoarder or privacy-conscious power user with a home server already running, it's a strong fit; for someone just wanting better recommendations on existing platforms, this isn't a drop-in replacement.
💰 Pricing
📊 Global score
🤖 AI-enriched data
100% gratuit, licence AGPL-3.0. Auto-hébergé, aucun coût d'abonnement — nécessite votre propre matériel (GPU 8GB VRAM min., 32GB RAM recommandés).
Pros
Traitement 100% local, aucune donnée envoyée à un tiers
Plusieurs modes de recherche complémentaires (fuzzy, embeddings, sémantique)
Identification 'soft hash' qui survit aux renommages/déplacements de fichiers
Indexation distribuée pour chercher une bibliothèque distante sans tout télécharger
Architecture modulaire (images, musique, texte, vidéo)
Cons
Configuration matérielle exigeante (GPU 8GB VRAM, 32GB RAM, 20-25GB disque)
Courbe d'apprentissage plus raide que les plateformes commerciales
Dépend de modèles ML externes à télécharger et gérer localement
Pas de synchronisation cloud ni de support multi-appareils natif