Skip to content

AI-assisted workflow

You can ask a coding assistant to draft a Datoviz example, then verify the result against the same examples and reference pages a human reader would use. This is especially useful while the direct Python API remains lower-level than a plotting library.

The canonical instructions and copyable prompt live on AI Agents Start Here. This page only explains how to make your request precise and how to review the result; it does not maintain a second copy of the agent contract.

Start in three steps

  1. Open AI Agents Start Here and copy its canonical prompt.
  2. Replace the task placeholder with the visual result, interaction, output, and language you want.
  3. Give the assistant access to datoviz.org so it can find an example and verify the current API.

For example, the task portion can be:

Task: show 50,000 two-dimensional points. Read positions and one scalar per point from NumPy
arrays, map the scalar to a perceptually uniform colormap, add XY panzoom, and save one 1600 x 900
PNG without opening a window. Return a complete Python program.

If the assistant cannot browse, provide these four inputs:

Ask the assistant to mark any API detail it could not verify instead of guessing.

Make the request specific

A useful request states the data, representation, interaction, output, and preferred language:

Instead of... Write...
"make a plot" "show 50,000 2D points colored by a scalar value"
"add controls" "let me pan and zoom the X/Y view with the mouse"
"make it 3D" "show 3D points with an arcball controller"
"export it" "render one frame to a PNG without opening a window"

Include array shapes and dtypes when they are already known. If they are not known, ask the assistant to state the contract it chose before the code.

Review the answer before running it

Check that the answer:

  • says whether the code is a complete program or an excerpt;
  • imports datoviz as dvz for the normal Python path, unless it explains a need for datoviz.raw;
  • uses a documented visual family and exact attribute names;
  • gives each array's dtype, shape, and item-count relationship;
  • creates the scene objects, attaches the visual to a panel, and opens a window or captures output;
  • checks feature and backend status rather than assuming native/WebGPU parity;
  • lists the example and reference pages used.

For Python, the normal v0.4 starting point is:

import datoviz as dvz

Datoviz v0.4 starts from scenes, visuals, and explicit data arrays. It does not provide the old high-level Python plotting API. If an answer invents Datoviz methods such as scatter() or imshow(), ask the assistant to restart from a current v0.4 example. High-level plotting belongs to GSP/VisPy2 and remains external work in progress.

Useful follow-up requests

After the assistant writes code, ask:

Which Datoviz documentation pages did you use, and which function or attribute names came from
each page?

For a larger result, also ask:

Separate the minimal verified Datoviz program from optional extensions. For every extension, state
its feature status and the example or reference page that supports it.

This keeps a working base visible when an optional backend or integration feature is uncertain.