Skip to content
CAVS Node

Browsing your data

The data plane stores content-addressed objects with no paths. The Git index puts human structure back on top of them: file trees, commit history, branches, tags and releases. This page is how to navigate it.

Files

Section titled “Files”

The Files tab renders the tree at a ref. Over the API:

Terminal window
export API=https://cavsnode.com/api/v1
export REPO=<repository-uuid>
export H="Authorization: Bearer $CAVS_TOKEN"


# Directory listing at the default branch
curl -sS "$API/repositories/$REPO/tree" -H "$H"


# A subdirectory on a specific ref
curl -sS "$API/repositories/$REPO/tree?ref=main&path=checkpoints/v3" -H "$H"


# One file's metadata (size, oid, last commit)
curl -sS "$API/repositories/$REPO/file?ref=main&path=model.safetensors" -H "$H"


# That file's history
curl -sS "$API/repositories/$REPO/file/history?path=model.safetensors&limit=50" -H "$H"
Parameter Applies to Default
ref tree, file the repository’s default branch
path tree, file, file/history repository root
limit file/history 50 (max 200)

Files that are LFS pointers show their oid, so you can jump straight to the artifact view or download the bytes.

Commits

Section titled “Commits”
Terminal window
# Recent commits
curl -sS "$API/repositories/$REPO/commits?limit=50" -H "$H"


# Only commits touching a path
curl -sS "$API/repositories/$REPO/commits?path=checkpoints&limit=20" -H "$H"


# Page backwards with an RFC 3339 cursor
curl -sS "$API/repositories/$REPO/commits?before=2026-07-01T00:00:00Z" -H "$H"


# One commit, with the artifacts it touched
curl -sS "$API/repositories/$REPO/commits/9a3f21c" -H "$H"

Each commit records which LFS artifacts it added or changed — that’s how the dashboard can answer “which commit introduced this 4 GB object?”.

Branches

Section titled “Branches”
Terminal window
curl -sS "$API/repositories/$REPO/branches" -H "$H"

Alongside each branch, CAVS reports health flags:

Flag Meaning
stale No commits in more than 30 days
oversized More than 2× the default branch’s storage footprint

Both are common sources of quiet storage growth — a long-lived experiment branch holding a dozen checkpoints nobody needs any more.

Artifacts

Section titled “Artifacts”

An artifact view is an object seen through the index: its oid, size, physical size, chunk count, download count, and the paths, commits and releases that reference it.

Terminal window
curl -sS "$API/repositories/$REPO/artifacts/$OID" -H "$H"
curl -sS "$API/repositories/$REPO/artifacts/$OID/timeline" -H "$H"

The timeline is where an artifact came from and where it went — first appearance, each commit that referenced it, each release that includes it.

The organization-wide registry

Section titled “The organization-wide registry”

Every artifact across every repository, in one filterable list:

Terminal window
curl -sS "$API/organizations/acme-ai/artifacts?sort=size&limit=20" -H "$H"
curl -sS "$API/organizations/acme-ai/artifacts?type=model&q=resnet" -H "$H"
curl -sS "$API/organizations/acme-ai/artifacts?repo=$REPO&sort=downloads" -H "$H"
Parameter Values
repo a repository UUID
type model, dataset, media, archive, other
q name substring
sort recent (default), size, downloads, name
limit page size
before RFC 3339 cursor; recent sort only

The response carries next_before — feed it back as before to page.

From the CLI:

Terminal window
cav artifacts --type model --query resnet
cav artifacts --org acme-ai --json | jq '.[] | select(.size > 1e9)'

Objects, without the index

Section titled “Objects, without the index”

If you want the raw data-plane view — no paths, no commits:

Terminal window
curl -sS "$API/repositories/$REPO/objects?sort=recent&limit=50&skip=0" -H "$H"
curl -sS "$API/repositories/$REPO/objects?sort=size&limit=20" -H "$H"

Each row is {oid, size, physical_size, chunk_count, created_at, downloads}. This works even on a repository that has never been indexed.

Section titled “Search”

Search spans repositories, files, commits and releases, scoped to the organizations you can see:

Terminal window
curl -sS "$API/search?q=resnet" -H "$H"
curl -sS "$API/search?q=checkpoint&type=files&ext=safetensors" -H "$H"
curl -sS "$API/search?q=fix+dataloader&type=commits&repo=$REPO" -H "$H"
curl -sS "$API/search?q=weights&type=files&min_size=1000000000" -H "$H"
Parameter Meaning
q the query (required)
type repos, files, commits, releases — omit for all
org restrict to one organization
repo restrict to one repository
ext file extension filter (files only)
min_size, max_size byte-size bounds (files only)

In the dashboard, press / anywhere to focus the search box. From the CLI: cav search "resnet".

Storage explorer

Section titled “Storage explorer”

Organization-level, for the “where did the terabytes go” question:

Terminal window
curl -sS "$API/organizations/acme-ai/usage/series?days=90" -H "$H"
curl -sS "$API/repositories/$REPO/storage/summary" -H "$H"
curl -sS "$API/repositories/$REPO/storage/folders?depth=2&limit=20" -H "$H"
curl -sS "$API/repositories/$REPO/storage/series?days=90" -H "$H"

storage/folders walks the indexed tree and aggregates by directory — depth between 1 and 5 controls how deep the roll-up goes. See Storage insights.

Next

Section titled “Next”