Source: https://docs.nygen.io/CyteBase/installation/
Revision: abe92a93ee6f586098196c484aad7221a7ce23e7
Structured page: https://docs.nygen.io/CyteBase/installation.json

Installation

CyteBase requires Python 3.13 or newer. Installing the cytebase package also installs CyteArc <https://docs.nygen.io/CyteArc/>, which opens and analyzes the datasets, DuckDB for catalog queries, and the Hugging Face client for bucket access.

Install with uv

Run these commands in the directory where you will work. uv <https://docs.astral.sh/uv/> creates an isolated .venv there:

uv venv --python 3.13
source .venv/bin/activate
uv pip install cytebase

In Windows PowerShell, activate the environment with .venv\Scripts\Activate.ps1 instead.

With pip, create and activate a virtual environment whose Python is version 3.13 or newer, then install the package:

python -m pip install cytebase

Note

CyteArc depends on hnswlib, which the installer compiles from source, so the installation needs a C++ compiler. The CyteArc installation guide <https://docs.nygen.io/CyteArc/installation> lists the compiler for each platform and a conda alternative for Windows.

Plotting

Plotting calls such as ds.plots.embedding use CyteArc’s plotting module, which needs CyteArc’s extra optional dependencies, including matplotlib and seaborn. The tutorials use them:

uv pip install cytebase "cytearc[extra]"

Without them, a plotting call raises an ImportError that names cytearc[extra].

Verify the installation

import cytebase defers loading its modules, so the check imports Catalog, which also loads CyteArc, and prints both versions:

python -c "from importlib.metadata import version; from cytebase import Catalog; print(version('cytebase'), version('cytearc'))"

To use the notebooks, install and launch JupyterLab from the same environment:

uv pip install jupyterlab
jupyter lab

Releases and buckets

Each public CyteBase release is a read-only Hugging Face bucket named by year and month, Nygen/cytebase-YYMM. A later release in the same month adds a patch number, as in Nygen/cytebase-2610.1. cytebase.releases() lists the public releases, newest first:

import cytebase

for release in cytebase.releases():
    print(release.version, release.bucket, release.released, release.latest)

Catalog() opens the release marked latest, currently 2610 in Nygen/cytebase-2610. Pin a release to reproduce discovery results:

catalog = cytebase.Catalog(version="2610")

To use another bucket, pass bucket= or set the CYTEBASE_BUCKET environment variable before starting Python. Both accept namespace/name or hf://buckets/namespace/name. The catalog takes its bucket from bucket=, then version=, then CYTEBASE_BUCKET, then the latest release. Passing both bucket= and version= raises an error.

Credentials

Public releases need no credentials. With the default token=None, CyteBase uses your Hugging Face credentials when they are available, such as an HF_TOKEN environment variable or a login with hf auth login, and reads anonymously otherwise. token=False always reads anonymously, and a token string reads with that token. A private bucket needs a token with access to it. Keep tokens and private bucket names out of notebook source and outputs.

Anonymous Hugging Face access allows 500 API requests per 5 minutes per IP address, and opening a store and reading its metadata uses a share of them. After a 429 Too Many Requests error, wait for the window to reset, or sign in with hf auth login for a higher limit.

Network access

CyteBase reads from Hugging Face over HTTPS. Behind a firewall or proxy, the Python process needs access to the Hugging Face Hub.

Catalog. Catalog() downloads the DuckDB catalog and keeps one verified copy at ~/.cytearc/cytebase.duckdb, or %LOCALAPPDATA%\cytearc\cytebase.duckdb on Windows. Every new query compares that copy with the published SHA-256 checksum and downloads the catalog again when it has changed, so queries need network access even when a copy is cached.

DataStores. Opened and mounted stores read metadata and count blocks from the bucket as needed. CyteBase keeps no persistent local copy of the counts.

Mounts. mount_datastore writes new analysis results to a local path, but its counts stay remote, so later reads also need network access.

Install from source

Clone the repository and let uv create the project environment:

git clone https://github.com/NygenAnalytics/CyteBase.git
cd CyteBase
uv sync

The checkout’s default environment also includes the test tools and the dependencies of the operator-only build pipeline <https://docs.nygen.io/CyteBase/developers/build-pipeline>.

Troubleshooting

Catalog(version=...) fails with an unexpected keyword argument: the code imported CyteArc 1.1’s earlier copy of the SDK, cytearc.cytebase. Import cytebase instead.

A plotting call raises an ImportError naming cytearc[extra]: install CyteArc’s plotting dependencies as described in Plotting <https://docs.nygen.io/CyteBase/installation/#installation-plotting>.

429 Too Many Requests: wait for the anonymous request window to reset, or sign in with hf auth login.

CyteBase imports in a terminal but not in a notebook: compare python -c "import sys; print(sys.executable)" with the notebook kernel, then launch Jupyter from the environment that contains CyteBase or select its kernel.

Next steps

Follow Explore CyteBase <https://docs.nygen.io/CyteBase/tutorials/cytebase> to search the catalog and plot a published dataset.