neuthek — the next best cloud storage solution

Your data isn't ours. We never train on it. We never sell it.

neuthek runs on a server you control (self-hosted) or in your own tenant fenced behind Postgres row-level security (managed hosted). Your photos, videos, documents, face embeddings, summaries, and search history are not exported to third parties, are not used to train AI models — ours or anyone else's — and are not sold to ad networks, brokers, or partners.

What we're building

• Search by meaning — CLIP-class embeddings, vector index, hybrid CLIP-cosine + Postgres-FTS ranker.
• Content-aware compression — LinUCB bandit picks per-image codec (WebP / MozJPEG / AVIF / JXL).
• Open source self-host (free) or managed hosted (waitlist).
• Stack: FastAPI, PostgreSQL, pgvector, Redis, MinIO, OpenCLIP, Florence-2.

Common questions

What is neuthek? An AI-aware personal cloud storage product in active development. Is it open source? Yes — the self-host build will be released under an open-source license. No committed date yet; hosted launches first. Does it train AI on my photos? No — your data is not used to train any model, and is never sold. How does AI search work? OpenCLIP ViT-L-14 embeds each photo into a 768-dim vector; queries embed into the same space and pgvector ranks by cosine similarity. What does it run on? FastAPI + PostgreSQL 16 + pgvector + Redis + MinIO + a separate ML worker. Self-host via docker-compose. See the full FAQ for more.