Datoviz — GPU visualization for large scientific data¶

Datoviz is an open-source, cross-platform visualization engine written in C, with direct Python/NumPy bindings. It uses Vulkan to render large, dynamic datasets in interactive, high-quality 2D and 3D scenes on Linux, macOS, and Windows.

Datoviz is desktop-first, with native windows, offscreen rendering, application embedding, and built-in GUI support through Dear ImGui. An experimental WebGPU/WASM subset brings selected scenes to the browser.

MIT license C API and Python/NumPy Linux, macOS, and Windows Desktop-first

v0.4.0rc1: the first public release candidate is available for testing. The native Vulkan scene API is the primary supported path; browser WebGPU and advanced/unstable facilities remain experimental. See release notes, feature status, or the legacy v0.3 documentation.

Interactive 3D protein visualization rendered with Datoviz

What can you build?¶

Datoviz combines scientific visuals, interaction, annotation, and native application controls in the same retained scene model.

Textured Earth with orbital debris in a star field

Textured planets

Textured worlds, a star field, orbital debris, and 3D navigation.

Point cloud

Large colorized LiDAR data with fly navigation and depth enhancement.

U.S. state population density shown as a choropleth map

U.S. state choropleth

Population density mapped across contiguous U.S. state polygons.

Human cortical activity

Time-varying MEG activity projected onto a rotating cortical surface.

An animated spiral galaxy rendered from colored particles

Density-wave galaxy

An animated spiral galaxy built from a density-wave particle model.

GPU-driven colored particles forming an animated smoke plume

GPU particle smoke

Compute-driven particles rendered through native compute and graphics interop.

Scientific Plotting Workflow

Histograms, uncertainty bands, stacked traces, axes, and annotations.

Allen mouse brain

An anatomical RGBA volume with slices and structure meshes.

Start with your language¶

Python + NumPy

Create retained scenes and pass NumPy arrays directly to Datoviz functions. The API keeps the same explicit dvz_* vocabulary as C.

Python quickstart → Python API

C or C++

Use the native C library for desktop applications, embedding, offscreen rendering, capture, and lower-level runtime integration. The public API uses C linkage and is callable from C++.

First C program → C/C++ integration

Prebuilt wheels containing the Python binding and native runtime are published and validated for supported Linux, macOS, and Windows targets. Install the exact release candidate with python -m pip install --pre datoviz==0.4.0rc1, then follow the installation instructions.

Built for scientific applications¶

Large, dynamic data¶

Update points, images, sampled fields, meshes, volumes, text, and annotations without rebuilding the entire scene.

First-class 2D and 3D¶

Compose axes, panels, paths, images, and guides alongside cameras, lighting, meshes, point clouds, and volumes.

Interaction and composition¶

Combine linked panels, panzoom and 3D controllers, picking, queries, labels, colorbars, and native GUI controls.

Native integration and output¶

Render in windows or offscreen, embed the C engine, capture screenshots, export video, and record or replay render streams.

Where Datoviz fits¶

Datoviz v0.4 is an explicit rendering engine, not a Matplotlib-like plotting frontend. Most users work with retained figures, panels, visuals, and controllers; experienced integrators can use the protocol and runtime layers directly.

High-level plotting Developing GSP and VisPy 2 interfaces outside Datoviz v0.4 — use the scene API below today
Retained scene API Figures, panels, visuals, controllers, annotations, and queries
Datoviz Rendering Protocol v2 (DRP2) Backend-neutral, WebGPU-shaped command streams, validation, and replay
Rendering runtimes Native Vulkan execution and the experimental browser WebGPU subset

Source builds expose switches for compatible core, Vulkan, canvas, DRP2, scene, app, and GUI components. The native WebGPU build option is separate from the browser WebGPU/WASM toolchain. See Choose your layer and Build options.

Platform and maturity¶

Surface	v0.4 position
Native Vulkan scene API	Supported Primary path, with feature-specific status documented separately
Linux, macOS, and Windows	Supported targets Wheel and source-build requirements vary by platform
Python	Supported Direct generated binding with documented NumPy adaptation
Dear ImGui	Supported Built-in native desktop GUI support
WebGPU/WASM	Experimental Browser subset for promoted examples, not native feature parity
DRP2 and lower-level runtime APIs	Advanced/unstable Integration surfaces

Review Platform support, Feature status, and Project status before adopting an experimental or backend-specific feature.

Continue¶

For human-led coding-assistant guidance, see the AI-assisted workflow. For autonomous code generation, use the stricter AI-agent contract. Project acknowledgements are recorded in the repository credits.