canvasxpress-dashboards (Python)

Linked, multi-panel CanvasXpress dashboards in Python — for Jupyter, VS Code and Quarto notebooks, Dash, Streamlit, Flask and standalone pages. It renders a canvasxpress-dashboards spec, or builds a whole dashboard from a DataFrame. The renderer is bundled; there are no required dependencies (pandas only for a DataFrame).

from canvasxpress_dashboards import Dashboard

# A table in, a linked dashboard out. A patient cohort (numeric time + 0/1 event columns) gets
# filters, a Kaplan-Meier curve, a measure by group and against time, and a case card; a per-sample
# sequencing run-QC table gets filters, depth, mapping vs duplication and quality by lane.
Dashboard(data=trial, id="patient")

# Or render a spec you wrote, saved from the builder, or returned by the MCP tool.
Dashboard("my-dashboard.spec.json")

Click a patient's point and its record loads into the case card; every panel shares the data, so a selection in one panel selects the same rows in the others. If no template fits the table the page says so rather than guessing — pass a spec instead.

Where it runs

Flask and Dash are covered by the package's browser tests. Streamlit (1.66) and Shiny for Python (1.8) were run by hand: each draws the whole dashboard, with every chart, inside the frame. The page is embedded as-is in a frame that runs JavaScript, so pass it only dashboards you trust.

Data

cxd_data(df, id="patient") is what Dashboard(data=df) uses: numeric columns are the measures, every other column an annotation of the records, rows are the records (more rows than numeric columns is the orientation the templates expect). A DataFrame, a dict of columns or a list of records all work; missing values become null.

Talking to the page

A dashboard rendered with frame=True (the notebook output does) reports to the page that embeds it with window.postMessage: {type: "cxd-params", id, params} whenever a parameter changes — for the case-card dashboards, params.case is the patient last clicked — and {type: "cxd-resize", id, height} as it grows. Dashboard.frame_id is the id.

Engine and bundle

The dashboards need the CanvasXpress engine, loaded from canvasxpress.org by default (the current release, which dashboards use features of); pass engine_url= for a mirror or an offline folder holding canvasXpress.min.js and canvasXpress.css. The dashboards bundle (1 MB) is inlined in each page; bundle_url= loads it from a URL instead, which keeps notebooks small.

Not included

The server-computed insights strip (insights in a spec) needs the dashboards server and is not shown. Connector, dataset and live sources need it too; inline data (what a DataFrame becomes) works anywhere.

MIT licensed.