Getting Started

Quick Start

Init a repo, commit a checkpoint, and inspect its history — all locally, in a few commands.

1. Install

Dits ships as a single binary. If you have not installed it yet, follow the installation guide (it is identical to the engine install), then confirm it is on your path.

dits --version

2. Initialize and commit a checkpoint

Create a repository in the directory that holds your artifact, stage the file, and commit it. The first commit is your baseline.

dits init
dits add model.safetensors
dits commit -m "baseline"

On add, Dits chunks the file with FastCDC and stores each unique chunk by its BLAKE3 hash. Re-committing a lightly-changed checkpoint only stores the chunks that actually differ.

3. Inspect status and history

After making changes — say, a new training run overwrites model.safetensors — check what moved, commit again, and review the log.

dits status
dits add model.safetensors
dits commit -m "epoch 10"
dits log

4. Diff and check out earlier versions

Compare any two points in history, then restore an earlier checkpoint by its commit id. Reconstruction is byte-exact.

dits diff
dits checkout <commit>

Note

Everything on this page works today and runs entirely on your machine. Networked sync — push, pull, and fetch — is on the roadmap, so for now share artifacts by copying the repository.

Next steps

Checkpoint workflow — a fuller loop for iterating on a model.
Addressing — why content hashes give you dedup and lineage for free.
CLI for AI and the full engine CLI reference.