pyheltycloud

Async Python client for the hcloud backend of Helty, used by the recent control panels of Helty Flow VMC units and by the "Helty Home" app.

Born as the foundation for a helty_cloud Home Assistant integration, next to helty which speaks the local protocol of the previous panels. The new panels expose no listening port: control necessarily goes through the manufacturer's cloud.

Unofficial. The API is undocumented: it has been reconstructed by observing the app and verified on real hardware. Helty may change it at any time.

Installation

pip install pyheltycloud

Usage

import asyncio
from pyheltycloud import HeltyCloud, VmcMode

async def main() -> None:
    async with HeltyCloud("me@example.com", "password") as helty:
        for device in await helty.get_devices():
            print(device.name, device.model)

            # update() wakes the board and then reads: the method to use for polling.
            state = await helty.update(device)
            print(state.mode, state.temperature_indoor, state.humidity)

            # set_mode_verified() re-reads the state and resends if the command is lost.
            await helty.set_mode_verified(device, VmcMode.SPEED_2)

asyncio.run(main())

How it works

Two separate channels. The panel runs no server: it holds an outbound connection to AWS IoT Core (eu-central-1), and Helty's backend relays to it. Clients never reach the panel directly — the app and this library call the REST API at api.hcloud.heltyair.com, authenticating against an AWS Cognito user pool with the account's own email and password (SRP, nothing to provision).

That shape explains both surprises below: a read returns the last message the panel pushed, not the machine's current state, and a write is a message handed to the relay, so a 200 only means the backend took it.

Things to know before using it

Read, do not poke. The board reports on its own every 5.5 minutes at the median and every 33 at the 99th percentile, measured over 5334 spontaneous reports, because it reports on any significant change as well as on its timer. Reading the last message costs one call and the board never notices, so a poll should use get_last_telemetry(). refresh() instead wakes the board, and update() wakes it and waits up to ~15 seconds for the answer: keep those for right after a command, when confirmation is worth the round trip. The vendor asks not to solicit on a short timer.

Commands are acknowledged, not guaranteed. The cloud answers 200 before the board has done anything. set_mode() is a single send; set_mode_verified() re-reads the state and retries, for callers that would rather pay a round trip than miss a command.

Two different serials. Reads use device.serial_number (the machine), writes use device.board_serial (the board). They are not interchangeable.

What is covered

Development

uv sync --group dev
uv run pytest --cov=pyheltycloud
uv run ruff check src tests && uv run ruff format --check src tests
uv run mypy

Tests never touch the network: they use a fake local HTTP backend.

License

MIT.