neon3-sdk

Python client SDK for Neon3 neon3.rpc and neon3.event.

Install

python -m pip install --upgrade neon3-sdk

Package page: https://pypi.org/project/neon3-sdk/

Application API

from neon3_sdk import NeonApp, ObservableStore

with NeonApp.start(mode="windowed", origin="my-app") as app:
    state = ObservableStore({"selected": None})
    app.ui.mount_flow_file("app.nui")
    app.intent("domain.select")(lambda event: state.value("selected").set(event.payload["key"]))
    app.run()

NeonApp manages runtime lifecycle, UI revisions, intent routing, and Store publication. Direct NeonClient remains available for protocol-level code.

Start Neon3

RuntimeSession starts the Neon3 services as separate processes. On Windows it resolves and downloads the latest runtime release from:

https://github.com/unco999/Neon3-CiJian/releases

from neon3_sdk import RuntimeConfig, RuntimeMode, RuntimeSession

with RuntimeSession(RuntimeConfig(mode=RuntimeMode.WINDOWED)):
    print("Neon3 services are running")

The resolved release is cached under %LOCALAPPDATA%\Neon3Sdk\runtime\<tag>. Set NEON3_RUNTIME_VERSION=<tag> (for example v0.2.3) when reproducible pinning is required; only pin a runtime that is at least as new as the UI schema your flows use, otherwise newer nodes such as tooltip or canvas fail with nui_flow_unknown_attribute. Set NEON_ROOT or pass RuntimeConfig(neon_root="D:/Neon3", profile="debug") to use a local checkout. The SDK starts neon-eventd, neon-wgpu-runtime, and neon-ui-runtime; it does not create windows or GPU resources itself.

RPC Usage

from neon3_sdk import NeonClient, UiClient

rpc = NeonClient.connect("127.0.0.1:39102", origin="my-tool")
ui = UiClient(rpc)
program = ui.submit_flow('version 1\nsurface example revision 1\nsurface root\n')
print(program.surface_id)

Event Usage

from neon3_sdk import EventClient

with EventClient.connect("127.0.0.1:39101").subscribe(
    name="ui.file_drop.accepted"
) as events:
    for image in events.file_drops():
        print(image.file_name, image.source_path)

ui.file_drop.accepted is the existing Neon3 event bridge for OS file drops. Image tools can use it to start OpenCV analysis without polling.

Tests

python -m unittest discover -s tests -v