cvh-client

Python API client for the Community Visualization Hub (CVH) — a platform for creating and sharing interactive genomic, spatial, and single-cell visualizations powered by Gosling.js and Vitessce. Designed for use in Jupyter notebooks.

Source lives in the CVH monorepo under python-client/. File issues at hms-dbmi/cvh/issues.

Installation

Requires Python 3.10+.

pip install cvh-client

With Vitessce widget support:

pip install "cvh-client[vitessce]"

Quick start

from cvh_client import CVHClient

# Authenticate against the hosted CVH via Auth0 (opens a browser
# for the device-authorization flow). Swap `base_url`, `domain`,
# `client_id`, and `audience` for your own deployment's values if
# you're self-hosting.
client = CVHClient.from_login(
    base_url="https://api.visualizationhub.org",
    domain="auth.visualizationhub.org",
    client_id="FKDEMc1rao9wVRnMIE9RznPsiLeHRXJS",
    audience="cvh-api-prod-id",
)

# List workspaces
workspaces = client.list_workspaces()
ws = workspaces.items[0]

# List datasets and visualizations in a workspace
datasets = client.list_datasets(ws.uuid)
visualizations = client.list_visualizations(ws.uuid)

# Get a single visualization with its config
viz = client.get_visualization(visualizations[0].uuid)

# Update a visualization, then publish it
client.update_visualization(viz.uuid, name="New name", conf={...})
client.publish_visualization(viz.uuid)

Token expiration is handled automatically — if the client was created via from_login(), it will re-authenticate when the token expires.

API methods

Workspaces

Datasets

Visualizations

Public

The visualization methods work without signing in, so CVHClient(base_url="https://api.visualizationhub.org") is enough. list_public_workspaces requires sign-in.

All methods return typed Pydantic models (Workspace, Dataset, Visualization, etc.).

Error handling

API errors raise descriptive exceptions:

from cvh_client import NotFoundError, AuthorizationError, CVHAPIError

try:
    client.get_visualization("nonexistent-uuid")
except NotFoundError as e:
    print(e.status_code)  # 404
    print(e.detail)       # {"detail": "..."}
except AuthorizationError:
    print("Check your permissions")
except CVHAPIError as e:
    print(f"API error {e.status_code}: {e}")

Examples

See the examples/ directory for Jupyter notebooks:

• getting_started.ipynb — Authentication, listing workspaces/datasets/visualizations, basic CRUD
• vitessce_workflow.ipynb — End-to-end Vitessce workflow: create a visualization with the Codeluppi et al. osmFISH config, retrieve it, view it with the Vitessce widget, modify and update

Development

Run these from inside python-client/ (checkout the monorepo first).

uv sync                                # Install dependencies
uv run pytest                          # Run tests
uv run ruff check src/ tests/          # Lint
uv run ruff format src/ tests/         # Format

Releasing

Published to PyPI via manual GitHub Actions dispatch. To cut a release:

1. Bump the version in VERSION.txt, pyproject.toml and src/cvh_client/__init__.py (the workflow refuses to publish if they differ), and move the [Unreleased] changelog entries under the new version.
2. Merge to main.
3. In the CVH repo's Actions tab → Publish cvh-client → Run workflow, entering the same version string as confirmation. Choose testpypi first to rehearse the release on TestPyPI, then run it again with pypi.

PyPI never accepts the same version twice, so a mistake can't be fixed by re-uploading; it needs a new version. That's what the TestPyPI rehearsal is for.

After publishing, the workflow installs the new version and smoke-tests it against the matching backend (dev for TestPyPI, production for PyPI).

The workflow uses trusted publishing (OIDC — no long-lived token). Trust is configured on each index as:

• Repository: hms-dbmi/cvh
• Workflow: publish-client.yml
• Environment: testpypi on TestPyPI, pypi on PyPI