query-viz

Render tabular query results as chart images. Rows in, PNG out.

query-viz does one thing: it turns a list of row dicts (exactly what a data warehouse driver such as BigQuery, Snowflake or Postgres hands back) into a chart image. No agent framework, no cloud SDK, no DataFrame requirement. That makes it a clean building block for an LLM agent tool, a report generator, or any pipeline that has already fetched data and wants a picture of it.

from query_viz import render_chart

png = render_chart(
    rows=[
        {"month": "Jan", "signups": 120},
        {"month": "Feb", "signups": 155},
        {"month": "Mar", "signups": 143},
    ],
    chart_type="bar",     # bar | line | pie | scatter
    x="month",
    y="signups",
    title="Signups by month",
)

with open("signups.png", "wb") as fh:
    fh.write(png)

Why a library

It was extracted from an ADK marketing agent's "chart this query" tool. The rendering is pure and reusable; the agent-specific parts (running a guarded query, uploading the image to Drive) stay in the agent as a thin wrapper. Same split as sql-guard: the reusable engine is a library, the framework glue is not.

API

render_chart(
    rows,                     # Sequence[Mapping[str, Any]]
    *,
    chart_type,               # "bar" | "line" | "pie" | "scatter"
    x,                        # category / x-axis column (pie slice labels)
    y=None,                   # numeric value column (required for every type)
    title="",
    dpi=120,
    figsize=(10.0, 6.0),
    palette=None,             # brand colours, applied in order
    color_map=None,           # permanent category-to-colour pins
    color_overrides=None,     # one-off pins for this chart; beat color_map
) -> bytes                    # PNG image bytes

• bar, line, scatter need both x and y.
• pie uses x for slice labels and y for slice sizes.
• Non-numeric y values coerce to 0.0 rather than erroring, so a stray None won't sink a whole chart.
• Raises ChartError (a ValueError) for an unsupported type, empty rows, or a missing column.
• Colour pins are keyed by the category value for bar and pie, and by the y column name for line and scatter, which draw a single series.

Design notes

• Headless. Forces matplotlib's Agg backend, so it renders on a server with no display.
• Pure. No I/O, no globals: deterministic given its inputs, and safe to call from a worker thread (asyncio.to_thread).
• Typed. Ships py.typed.

Install

pip install query-viz

The distribution is named query-viz; the import package is query_viz. Only dependency is matplotlib.

License

Apache-2.0. See LICENSE.