GraphExec is a tool for executing litegraph.js graphs on a server, with node operations implemented as Python functions.

Slot types can be any Python types, including NumPy arrays and custom, opaque classes.

It runs as a REST back-end for litegraph.js with some aftermarket modifications, or as a command-line tool. Generating JavaScript code for client-side stub nodes (with slots, properties and widgets, but no business logic) is also possible, based on decorators added to the node implementation.

For each node, server-side code can return image results (including Matplotlib figures).

The project is in an alpha development stage; for example, litegraph.js built-in nodes are not implemented, and there is no support for the REST back-end in the upstream library.

With my_node_types.py:

from graphexec import node_type

from .my_implementation import _perlin_noise

@node_type.define("my_package/perlin_noise", "Perlin noise")
@node_type.input("width", "number")
@node_type.input("height", "number")
@node_type.output("map", "numpy.ndarray")
def perlin_noise(width, height):
    return dict(map=_perlin_noise(int(width), int(height)))

NODE_TYPES = [
    perlin_noise,
]

graphexec-generate-js-module -o ../litegraph/src/nodes/my-package-generated.js my_node_types
env PYTHONPATH=. graphexec-server --wwwroot ../litegraph -m my_node_types

black --diff **.py

mypy graphexec