Artifact Graph & Evaluate

oci-bai is an OCI-compatible artifact graph CLI and dashboard. Every image pushed through the gateway creates a commit in the graph — recording the manifest, files, derivation edges to parent images, and the full audit trail. The graph is the source of truth; the backing registry is just storage.

Full documentation: oci.briefcaseai.io

The commit model

A commit is defined by four things:

#	Part	What it is
1	Ref	A tag or digest that names this version (e.g. `v1`, `cartpole`)
2	Family	The repository the image belongs to (e.g. `rl-gym-env`)
3	Derivation	The parent commit(s) inferred from shared layers, or declared in a cohort push
4	Manifest	The full OCI manifest, files, and SBOM recorded at push time

Lifecycle

Push through the gateway

Tag and push with any OCI-compatible tool. Repositories are created automatically.
Terminal window
```
docker tag my-image:latest localhost:8080/my-repo:v1
docker push localhost:8080/my-repo:v1
```
Confirm the commit
Terminal window
```
oci-bai --repo my-repo log v1
```

Inspect provenance

oci-bai --repo my-repo provenance v2
oci-bai --repo my-repo whodepends base-v1

Compare versions

oci-bai --repo my-repo diff base v1 --depth package
oci-bai --repo my-repo diff base v1 --depth bench

Search the catalog

oci-bai search "format==safetensors cuda>=12.4"
oci-bai search "gymnasium==1.1.0 arch==arm64"

Run modes for evaluation

Two commands support active evaluation:

Command	What it does
`oci-bai attach-bench <ref>`	Attach a benchmark verdict to a commit
`oci-bai --repo <r> diff <base> <candidate> --depth bench`	Compare verdict deltas between two versions
`oci-bai hunt <ref>`	Host a verdictml HuntEnv episode against the candidate

Where this fits

Quick start Push your first image and run your first search in under five minutes.

CLI reference Every oci-bai command and its flags.

Full docs The complete oci-bai documentation at oci.briefcaseai.io.