Semantic Unison is an LLM-friendly fork of Unison with semantic extensions for efficient, trustworthy file synchronization. AI-assisted code, documentation, and contributions are welcome, with transparent attribution, reviewable changes, and testing.
Our goals:
- Semantic hashing: distinguish image payloads, standard metadata, previews, and private fields while preserving unknown bytes exactly.
- Gradual camera understanding: grow DNG and MakerNote coverage through composable, versioned plugins and incremental refinement of component hashes.
- Precise updates and merges: transfer bounded metadata changes and merge compatible edits while verifying the exact resulting file.
- APFS clone updates: reuse the receiver’s existing data in a private clone, reducing unnecessary rewriting for ordinary files as well as photographs.
- Optional Adobe validation: check DNG structure, decoding, and raw digests through explicitly configured validation policies.
- Coordinated parallelism: use OCaml and OxCaml workers efficiently, with explicit ownership, cancellation, and shared resource budgets.
We credit Unison’s authors and retain the project’s existing licensing and attribution. This fork is maintained independently of upstream.
Read the design and I am an honest working LLM.