Real-time signal processing, built on Python.

from scipy.signal import welch

def _on_input(self, values):
    # A plain NumPy array, no wrapper
    _, power = welch(values[0], fs=self.fs)
    self._output.dispatch([power])

Bring your Python code as it is

Nodes are ordinary Python classes, and data reaches them as NumPy arrays with no wrapper type to unpack. Hand it straight to SciPy, scikit-learn, or PyTorch, and turn research code into a working product.

Process signals in real time

Signals flow as batches of samples, so a node can handle a whole batch in one NumPy operation. With no packets or tables to build for each chunk, overhead stays low enough for live audio and sensor streams.

Run it from desktop to device

Build on Windows or macOS, then run your graphs on a Raspberry Pi, or on-device inside iOS and Android apps.

Your app ships with the workspace

Put a web page in the workspace's app folder, and the engine serves it next to the editor and the graph API. There is no server to set up or deploy, and the app travels with the workspace when you share it. New workspaces start with a working app template.

"nodes": {
  "Clock_1": { "type": "Clock" },
  "Wave_1": { "type": "WaveGenerator" }
},
"edges": [{ "source": "Clock_1.times",
            "target": "Wave_1.times" }]

Graphs are text, for people and AI

Each graph is a readable JSON file of nodes and edges, kept apart from the editor's layout. Change it in the visual editor, by hand, or with an AI coding agent, and review every change in Git. New workspaces include an AGENTS.md guide, and the engine can export its full node catalog as JSON.