phosphor 0.3

WebGPU accelerated visualization of unbounded neural data streams in Python

phosphor

GPU-accelerated real-time sweep renderer for multichannel timeseries data. Built on WebGPU and Qt, phosphor renders thousands of channels at high sample rates with minimal CPU overhead.

Designed for neuroscience and real-time signal monitoring -- push (n_samples, n_channels) numpy arrays and phosphor handles downsampling, autoscaling, and rendering.

Installation

pip install phosphor

Quick Start

import numpy as np
from PySide6.QtWidgets import QApplication
from phosphor import SweepConfig, SweepWidget

app = QApplication([])

widget = SweepWidget(SweepConfig(
    n_channels=128,
    srate=30000.0,
    display_dur=2.0,
    n_visible=64,
))
widget.show()

# Push data from any source -- shape: (n_samples, n_channels), float32
widget.push_data(np.random.randn(500, 128).astype(np.float32))

app.exec()

Embedding in an Existing Qt Application

SweepWidget is a standard QWidget that can be added to any layout:

from PySide6.QtWidgets import QMainWindow, QVBoxLayout, QWidget
from phosphor import SweepConfig, SweepWidget

class MyWindow(QMainWindow):
    def __init__(self):
        super().__init__()
        self.sweep = SweepWidget(SweepConfig(n_channels=64, srate=1000.0))
        self.setCentralWidget(self.sweep)

    def on_new_data(self, data):
        self.sweep.push_data(data)

Runtime Configuration

Update parameters without recreating the widget:

from phosphor import SweepConfig

widget.update_config(SweepConfig(
    n_channels=256,
    srate=30000.0,
    display_dur=4.0,
    n_visible=128,
))

Built-in Demo

python -m phosphor
python -m phosphor --channels 256 --srate 30000 --visible 64 --dur 2.0

Keyboard Controls

Key	Action
Up / Down	Scroll channels by 1
Page Up / Page Down	Scroll channels by one page
[ / ]	Halve / double visible channel count
- / =	Y-axis zoom out / in (disables autoscale)
A	Toggle autoscale
, / .	Halve / double display duration

Future: Migration to fastplotlib

Phosphor currently uses custom WGSL shaders and raw wgpu-py for rendering. The plan is to migrate to fastplotlib (built on pygfx) once both libraries reach 1.0 (targeting mid-2026), which would eliminate the custom shader maintenance burden. The migration can happen in stages:

1. Keep SweepBuffer -- the CPU-side circular buffer with incremental min/max downsampling is the core of phosphor's performance story. Neither pygfx nor fastplotlib provide built-in min/max downsampling for line data, so this logic stays.
2. Replace GPURenderer + WGSL shaders with fastplotlib's LineStack, feeding it the already-downsampled min/max columns from SweepBuffer. This drops the custom pipeline/shader code and gains fastplotlib's built-in axes, colormapping, and interaction tools.
3. Evaluate whether ChannelPlotWidget can be simplified or replaced by fastplotlib's Figure/Subplot layout, retaining the keyboard controls and channel pagination UX.

Development

We use uv for development.

1. Fork and clone the repository
2. uv sync to create a virtual environment and install dependencies
3. uv run pre-commit install to set up linting and formatting hooks
4. uv run pytest tests to run the test suite
5. Submit a PR against the dev branch