SDK and CLI

The systemonemodels package installs the systemone command and a small Python library over the same HTTP API the website uses. It needs Python 3.10 or newer. It is open source — systemonemodels/systemonemodels-sdk on GitHub, Apache-2.0 — so you can read exactly what it does with your token.

pip install systemonemodels

Sign in

systemone login

The CLI shows an eight-letter code and opens systemonemodels.ai/device in your browser. Check the code matches, choose Authorize, and the terminal finishes on its own. The token it receives is listed under Settings → Access tokens as "CLI · your-machine", where you can revoke it.

Over SSH, or anywhere a browser cannot open, the CLI prints the address instead — open it on your laptop or phone and type the code. --no-browser does the same on purpose.

For CI and scripts, skip the browser and use a token created in settings:

systemone login --token "$SYSTEMONE_TOKEN"   # store it
export SYSTEMONE_TOKEN=s1_pat_...            # or just set it; nothing is stored
systemone whoami

The stored token lives in your config directory with 0600 permissions. systemone logout forgets it.

Discover

systemone search snake
systemone search --capability route --architecture laya --sort downloads
systemone show biplov/snake-balanced-multilingual

--sort takes relevance, recent, stars, downloads, accuracy or latency.

Pull

systemone pull biplov/snake-balanced-multilingual --variant onnx-int8
systemone pull biplov/snake-balanced-multilingual --version 0.1.0 --dest ./snake

Files are cached by content, so a second pull downloads nothing and two variants that share a tokenizer store it once. --dest links the tree into a folder of your choosing instead of printing the cache path. systemone cache info shows where the cache is and how big it has grown; systemone cache clear empties it.

Publish

cd ~/my-workspace
systemone push

Run push in a folder and it finds every model underneath — trained checkpoints and their ONNX and Core ML exports, GGUF files, any folder holding model.onnx, model.safetensors, a .gguf or an .mlpackage — lists them, and asks which to publish (1,3-5 or all). Variants of one model are published together, as one version with a folder each. To publish one folder under a name of your choosing:

systemone push ./runs/snake/model --repo you/laya-snake

Nothing has to be retyped:

• The folder's README.md becomes the model card, and its Hugging Face front matter sets the licence, tags and base model. A model without one gets a card written from its evaluation, questions and variants.
• The systemone.yaml manifest is inferred: capabilities from the question schema; accuracy, calibration error and latency from a System One Studio run's eval.json or an export's own measurements; and the base model, linked to Hugging Face when the files record that it came from there. --manifest uses yours instead.
• The version is the next minor after the latest (0.1.0, 0.2.0, …), or --version.

push shows its plan before sending anything and, in a terminal, asks first — unless you named the repository with --repo or passed --yes. --dry-run prints the manifests, cards and files, then stops. Files already in the registry are not uploaded again, so republishing with one changed file sends one file. Other options: --namespace publishes under an organization, and --license, --readme, --variant, --notes and --private do what they say. --all --yes publishes everything found without a question, for scripts.

systemone create you/name makes an empty repository, and systemone validate systemone.yaml runs the registry's own manifest check, so a manifest that passes locally will not be rejected on push. Publishing needs a verified email address. Folder discovery needs systemonemodels 0.2.0 or later.

In Python

from systemone import snapshot_download

path = snapshot_download("biplov/snake-balanced-multilingual", variant="onnx-int8")
# path is a local directory with the version's files, served from the cache
# on every call after the first.

Running a model is up to the runtime it was exported for — ONNX Runtime, Core ML and so on. The repository's model card says which, and show prints its runtime.