One engine, two products
The hard, valuable part of Dits isn't video version control or weight version control — it's the content-addressed engine underneath. The same primitives win in two very different markets.
The shared engine
Both products sit on the same foundation: content-defined chunking (FastCDC), BLAKE3 content addressing, a deduplicating content-addressed store, and delta sync. Each product is mostly a format transform plus a workflow skin — media plugs in keyframe-aware video handling; AI plugs in tensor-aware layouts and embedding fingerprints.
The roadmap that makes both genuinely better than what exists is the same for both: move past exact-match hashing to similarity-addressing (dedupe near-duplicates) and derivation-addressing (store the recipe, recompute the artifact). Anyone can build “Git for video.” Almost nobody can build one engine whose primitives win in two markets at once.
Dits is open core and local-first. The engine, formats, and protocol are inspectable and self-hostable; the optional cloud layer adds managed compute for recompute-heavy work. Your weights and footage never have to leave your infrastructure.
Follow the build
Dits is built in the open. Watch the engine, the AI profile, and the sync layer come together.