Getting started

1. Install an SDK

Node Data ships first-party SDKs for the four languages most common in robotics and applied ML. The Python SDK is the reference implementation; every endpoint surfaces there first.

# Python
pip install nodedata

# TypeScript / Node.js
npm install @nodedata/sdk

# Go
go get github.com/nodedata/go-sdk

# Rust
cargo add nodedata

2. Create an API key

Sign in to Dashboard → API keys and generate a key. Keys are scoped to either test (sandbox marketplace, no real funds) or live (production). Export the key as an environment variable:

export NODE_DATA_API_KEY="nd_live_•••••••••••••••••••••"

3. Make your first call

Search the marketplace and print the first five hits. No payment is required to call the search API.

from nodedata import NodeData

node = NodeData()  # reads NODE_DATA_API_KEY from env

# Search the marketplace
results = node.search("manipulation policy")
for hit in results.items[:5]:
    print(hit.slug, hit.price, hit.license)

4. Download a model

Downloads stream from a region-aware CDN with resumable byte ranges. Models you have not licensed can still be inspected via their model card, but the weight files require a successful checkout.

model = node.models.retrieve("acme/grasp-policy-v2")

# Stream weights into local cache
path = node.models.download(
    model.id,
    revision="1.4.0",
    destination="~/.cache/nodedata",
)
print(f"Saved to {path}")

5. Publish a model

Uploading is symmetric: hand the SDK a list of file paths, a license, and pricing. The SDK chunks large files via the resumable upload protocol and writes a content-addressed manifest.

node.models.create(
    name="lift-and-place",
    framework="pytorch",
    task="manipulation",
    license="MIT",
    files=["./weights.safetensors", "./model_card.md", "./config.json"],
    price_usd=49.00,
)

Next steps

• Uploading models — packaging, formats, and validation.
• API reference — every endpoint documented.
• Deploying models — Jetson, ROS 2, and cloud inference.