Skip to content
CAVS Node

CAVS Node documentation

Storage, versioning and distribution for large files — models, datasets, media, images, game and build artifacts. Fully Git LFS compatible, and every transfer moves only what actually changed.
QuickstartWhat is CAVS Node?API reference

Get moving in three commands

Section titled “Get moving in three commands”

  1. Install the CLI and sign in.

    Terminal window
    curl -fsSL https://raw.githubusercontent.com/orelvis15/cavs-hub-cli/main/install.sh | sh
    cav config set api https://cavsnode.com
    cav login

  2. Connect a Git repository to a CAVS repository you created in the dashboard.

    Terminal window
    cav repo connect my-org/my-repo

  3. Push as usual. Large files go through CAVS; Git keeps the history.

    Terminal window
    git add model.safetensors
    git commit -m "Add trained model"
    git push

Full walkthrough in the Quickstart.

Pick your path

Section titled “Pick your path”
Git-native workflowKeep using git add / commit / push. CAVS replaces Git LFS storage and transfer with chunk-level dedup.
Python SDKUpload and download artifacts from training jobs and pipelines with verified content hashes.
TypeScript SDKNode 18+ client with streaming uploads and downloads for build and release tooling.
CI/CDGitHub Actions, GitLab CI and Jenkins recipes for publishing and consuming artifacts.
Plain HTTPNo SDK for your language? The upload and download choreography is four curl calls.
Self-hostingRun the whole control plane, data plane and workers yourself, on your own S3 bucket.

What makes it different

Section titled “What makes it different”

Content-addressed, chunk-level dedup

Files are split with content-defined chunking and compressed. Consecutive versions of a large file share nearly all their chunks, so a 22 MiB file with 3 MiB of real change stores and transfers about 3 MiB — not 22 MiB.

Git LFS compatible, two ways

Point lfs.url at your repository and stock git-lfs works with zero extra software. Install the CAVS transfer agent and the same repository gains chunk-level dedup on both push and pull.

Bytes never touch the control plane

Uploads and downloads run against presigned object-storage URLs. The API authorizes once, then gets out of the way.

Your bucket, if you want it

Bring Your Own S3 keeps the bytes in your AWS, R2, Wasabi, B2 or MinIO account under your credentials and policies. CAVS holds only metadata.

Verified end to end

Every object is addressed by the SHA-256 of its bytes. Downloads that don’t hash back to their oid are rejected, never written.

Storage you can actually see

Per-repository logical vs physical bytes, dedup ratio, per-folder breakdowns, measured push/pull benchmarks, health checks and recommendations.

Reference at a glance

Section titled “Reference at a glance”
Dashboard cavsnode.com
API base URL https://cavsnode.com/api/v1
Health check https://cavsnode.com/readyz
Auth header Authorization: Bearer <token>
Content address lowercase hex SHA-256 of the object bytes (oid)
Python package pip install cavs
npm package npm i @cavsnode/cavs-sdk
CLI cav — install