Energy Node

Energy monitoring and automation for a remote site. It runs on a Raspberry Pi 1 and works with or without Home Assistant.

What it is

Some solar arrays, batteries and switchable loads sit at a workshop, barn, garage or second property. Those sites are often too far away, or sit behind too flaky a link, for the Home Assistant at home to poll their devices directly.

Energy Node puts one small Linux box at that site and makes it a self-contained node. It polls the local devices without any cloud account, runs its own automation rules and serves its own web dashboard. If the internet or the VPN goes down, the node carries on as before.

If you use Home Assistant, a Mosquitto bridge over Tailscale mirrors the node's devices to your main site. There they appear as ordinary MQTT Discovery entities: PV production, battery state of charge, load power, grid import/export. Without Home Assistant, the dashboard works on its own.

Highlights

• Live overview: energy flow, self-sufficiency ring, balance and cards you can arrange in a layout editor. You can pin favourite entities and give each device its own icon.
• History in the browser: charts are recorded in IndexedDB, so the Pi stores nothing. You can export them as CSV or JSON.
• Automations: a rule engine with conditions, hysteresis, hold times and cooldowns, edited in the dashboard. It runs as its own process, separate from the dashboard.
• Configuration from the UI: every device service is set up through a form in the dashboard, including Tuya and Tailscale setup wizards.
• Energy roles and device map: you tell the node which measurement is PV, battery, load or grid, and a map you arrange by hand records how the site is wired.
• Desktop installer: sets up a fresh Pi, updates an existing node or diagnoses and repairs one, from Windows, macOS or Linux.
• Updates from the dashboard: a notice appears when a new release is out. The dashboard downloads the signed package and redeploys the node without SSH.
• Home Assistant, two ways: devices arrive through MQTT Discovery, and three HACS integrations bring dashboard features into HA directly.
• Built for weak hardware: the dashboard is a Go program that needs no runtime or database on the node, so it starts fast and leaves most of the RAM to the services. Charts are drawn in your browser and nothing loads from a CDN, so the node also works without internet. The Raspberry Pi 1 (ARMv6, single core, 512 MB RAM) is the floor every design decision is measured against.

Every page of the dashboard, including all settings tabs and the four colour schemes, has a screenshot in the dashboard, page by page. The screenshots come from a local smoke test against a fixture broker, so you can reproduce them without a Pi.

Supported devices

All devices are polled locally, and none of them needs a cloud account.

Hardware	How	Direction
Shelly Gen1 and Gen2+ (switches, meters, 3-phase, H&T, ADC)	HTTP /status / /rpc	read + switch
APsystems EZ1 microinverters	local REST API	read + power limit, on/off
Lumentree inverters with a Trucki stick (T2SG/T2MG/T2HG)	HTTP	read-only
Tuya devices	local protocol via tinytuya	read + switch
LiFePO4 battery banks	computed from other devices (coulomb counting)	state of charge
The node itself	CPU, RAM, disk, throttling, Mosquitto/Tailscale status, pending updates	read

Each family is one small Python service. Device services covers what each one publishes. A new device family is added as another service; CONTRIBUTING.md explains how.

How it fits together

flowchart LR
    subgraph Remote["Remote site: Raspberry Pi 1 (ARMv6)"]
        DEV["Shelly · Tuya · APsystems EZ1<br/>Trucki stick · LiFePO4 banks"]
        BR["Python bridges<br/>(one systemd service each)"]
        MQ{{"Mosquitto<br/>localhost:1883"}}
        AUT["Automation service"]
        DASH["Go dashboard<br/>:8080 / Caddy :443"]
        DEV --> BR --> MQ
        MQ <--> AUT
        MQ --> DASH
    end

    MQ <-->|"outstation/# over Tailscale (optional)"| HA

    subgraph Main["Main site"]
        HA["Home Assistant<br/>+ Mosquitto"]
    end

• Local autonomy. The broker, services, automations and dashboard all run on the node. The link to the main site is optional.
• One broker, one narrow link. Everything on the node talks through a single local MQTT broker, and no service calls another over HTTP. Only one topic tree, outstation/#, crosses the Tailscale tunnel. That means no port forwarding and no exposed broker.
• Cheap hardware is the point. A side building usually gets a spare Pi 1 or a Pi Zero W. That is why there is no SQLite, no server-side history and no plugin system. Any newer Pi has room to spare.

This is the beginning of the README. Read the rest at the original. Read the full README ↗