Up to 1280 entities across fifteen device families. Real-time MQTT streaming, a local Modbus/TCP option for the three-phase PowerOcean, native Energy Dashboard integration, and control wherever the device allows it.
Through the official Developer API, the integration polls about every 30 seconds. Too coarse for meaningful automations or real-time energy monitoring. On my PowerOcean, the cloud copy behind that poll refreshed reliably only while the EcoFlow app or portal was open. One other owner has reported Standard Mode readings that moved only twice in one day.
EcoFlow Energy offers three connection types. Standard Mode uses the official IoT API with HTTP polling and real-time MQTT push. Enhanced Mode receives what the device pushes over the account connection, every 2 to 4 seconds for most devices, with no Developer API keys required.
My own PowerOcean now runs without the cloud. If EcoFlow support has enabled Modbus/TCP on a three-phase PowerOcean, EcoFlow Energy reads it directly on your network about every 2 seconds, with no account. It shows the main readings, not everything the account connection does.
If the connection drops, a 4-tier reconnect strategy kicks in automatically. In Enhanced Mode, entities degrade gracefully through defined stages while retaining last-known-good values. Data collection resumes without manual intervention. Set it up once.
Auto-discovery, three connection types, and sensors that map directly to the HA Energy Dashboard. No YAML required.
Enhanced Mode streams device pushes every 2 to 4 seconds for most devices. Delta and Smart Plug also get MQTT push in Standard Mode. If EcoFlow support has enabled Modbus, a three-phase PowerOcean can be read locally every 2 seconds, with no cloud.
All energy sensors use total_increasing state class. Solar, grid, battery, and home consumption map directly into the HA Energy Dashboard.
Enter your EcoFlow account or API keys once. All devices bound to your account are discovered automatically and managed within a single integration instance. A Local entry holds exactly one PowerOcean.
Toggle outputs, set charge limits, and configure SoC thresholds. Switches, selects, numbers, and buttons across nine of fifteen device families.
4-tier MQTT reconnect strategy with graduated availability degradation. If the real-time stream drops, entities retain last-known-good values while the connection recovers through defined stages.
Download HA diagnostics with full device state. Credentials are automatically redacted. Debug with confidence.
Home batteries, hybrid inverters, plug-in and whole-home batteries, load panels, portable power stations, an air conditioner, a wallbox, a plug, a meter, a microinverter, and a solar tracker. 1280 entities total, including per-pack, per-module and per-circuit entities up to each device's maximum. Multi-device configurations supported natively.
Home Battery System
Hybrid Inverter
Plug-in / Balcony Battery
Whole-home AC-coupled Battery
Portable Power Station
Portable Power Station
Smart Energy Plug
Portable Air Conditioner
EV Wallbox
Sensors, no controls
Sensors, no controls
Sensors, no controls
Three automations running in production. Each uses entity data from this integration, standard Home Assistant triggers, and zero custom code.
Solar Cascade Charging: three device types, one integration, zero grid export.
PowerOcean hits 98% SoC. Smart Plug turns on. Delta 2 Max charges from solar surplus instead of exporting to the grid. Plug turns off when Delta reaches 99% or PowerOcean drops below 50%.
Three device types coordinated through one integration. No solar energy wasted on feed-in.
Delta 2 Max powers office equipment via AC output. When SoC drops below 20%, a push notification fires. Solar surplus charges it back to 80% via max_charge_soc.
AC output off at 23:00, on at 07:00. Standby timeout configured through the number entity. Set it, forget it.
PowerOcean grid export exceeds 1000W for 5 minutes. Push notification with current export value. Shift the dishwasher, laundry, or EV charger into the solar window.
Not automation for its own sake. Actionable awareness that turns monitoring into lower grid bills.
A campsite pillar is fused anywhere from 2 to 16 amps, and a caravan running an air conditioner alongside a kettle can ask for more than that on one circuit. Three control layers sit between the two on my caravan, acting in order, so the pillar sees a steady draw instead of a spike.
Shore power control chain: the plug enforces the limit, the battery buffers the load, the fridge keeps running.
Current and power limits follow the hookup size in use. A charge power setpoint is a target for the battery, not a ceiling on the wire. At a 500 W setpoint with the air conditioner running, the grid pull measured 1415 W.
The plug is set to trip before the pillar's breaker sees that number. The Delta is already in bypass mode, so fridge and lights keep running from the battery while the switch resets from Home Assistant instead of a walk to the reception.
Every two minutes the regulation subtracts the caravan's own load and a reserve from the budget, then writes what is left to the Delta's charge power.
It reads a ten-minute rolling average rather than the instant number, so an air conditioner compressor cycling on and off looks like one steady draw. None of it has any effect unless the Delta's own Charge Speed switch sits on ADJUST.
Below 30 percent Home Assistant sheds the second AC outlet and sends a notification. Below 20 percent the Delta marks that same outlet non-essential and sheds it on its own, no network required.
Below 10 percent the discharge limit takes over on the device itself. The fridge, the cooler, and the air conditioner on the first outlet are never shed by the automation, only flagged.
This is the same integration behind the automations above, on my own caravan, against a harder constraint: a fuse that trips once and stays tripped until someone finds the reset. The full package, the YAML, and the two approaches that did not work are in the repository.
Caravan guide on GitHubStandard Mode uses the official EcoFlow IoT API with HTTP polling and MQTT push. Enhanced Mode receives device pushes over the account connection, every 2 to 4 seconds for most devices, with graduated availability degradation instead of HTTP fallback. Local reads a three-phase PowerOcean over Modbus/TCP on your own network.
| Aspect | Standard | Enhanced | Local |
|---|---|---|---|
| Connection | EcoFlow cloud, HTTPS polling + MQTT | EcoFlow cloud, WSS MQTT | Your network, Modbus/TCP, no cloud |
| Credentials | Access Key + Secret Key | EcoFlow email + password | None. EcoFlow support must enable Modbus on the inverter first |
| Update rate | ~30s polling, plus MQTT push for Delta and Smart Plug | ~2-4s for most devices | ~2s polling |
| Devices | Models the Developer API exposes, the only mode for PowerStream | All except PowerStream, required for the Enhanced-only models | Three-phase PowerOcean only |
| PowerOcean controls | Read-only sensors | Backup Reserve, Solar Surplus Threshold, Work Mode | Backup Reserve, Indicator Brightness, and a Modbus Control switch that locks the EcoFlow app while on |
| Stability | Official EcoFlow API | Community-driven, may break with EcoFlow updates | Measured on one unit, use at your own risk |
| Best for | Long-term operation on the official API | Real-time monitoring, fast automations, PowerOcean control | Owners who want no dependence on EcoFlow's servers |
Local is deliberately narrow. Modbus Control is always off after a restart, and the app gets control back about a minute after it is switched off. The integration writes only the keep-alive signal, Backup Reserve and Indicator Brightness. After each change it reads Backup Reserve and Indicator Brightness back from the inverter. Per-pack charge levels, grid phase values and other writes are not included.
A Local entry holds exactly one PowerOcean, and Reconfigure moves a single-PowerOcean entry between cloud and Local. Compared with the account connection over about 19 hours on one unit, the charge level was within 1 percent.
Local is an independent implementation, not provided or reviewed by EcoFlow. Thanks to @jensfr1, who showed how EcoFlow enables Modbus on the inverter.
What the integration delivers across data, control, and failure handling.
| Feature | What you get |
|---|---|
| Data source | MQTT push (Enhanced), local Modbus/TCP for the three-phase PowerOcean, graduated availability with last-known-good data |
| Cloud dependency | Cloud by default, with a local connection for the three-phase PowerOcean that needs no account |
| Connection resilience | 4-tier reconnect, graduated availability degradation |
| Control | Switches, charge speed, SoC limits, thermostat card, wallbox Start/Stop, optimistic lock, read-back on wallbox, heating rod, Backup Reserve and Indicator Brightness |
| Entity depth | Up to 308 (PowerOcean, plus 15 Local-only), up to 245 (Ocean 2), up to 179 (Panel 40), 113 (Delta 2 Max), 76 (Stream AC 5000), up to 70 (Delta 3), 68 (Stream), 38 (WAVE 3), up to 36 (Smart Home Panel 2), 34 (PowerPulse 2), up to 31 (DELTA Pro Ultra), 25 (PowerStream), 21 (Smart Meter), 15 (Smart Plug), 6 (Solar Tracker) |
| Mobile tolerance | Offline = expected state, not an error |
Depth compounds. An integration that handles one device cleanly is useful. One that handles fifteen device families, connection failures, and a van-mounted battery going offline for three days takes deliberate architecture.
Install via HACS, choose a connection type, and your devices appear automatically.
Open HACS in Home Assistant. Search for EcoFlow Energy, download, and restart.
Add the integration in Settings. Choose Standard, Enhanced or Local and enter the matching credentials, none for Local.
Devices are auto-discovered. Entities appear in your dashboards, Energy Dashboard sensors are ready to map.
A focused stack for reliability. Reverse-engineered protocols, typed Python, and a test suite that covers every parser.
EcoFlow Energy was built by one person with a clear division of responsibility: architectural decisions, domain knowledge, and protocol understanding stay with the human. Test generation, code review, schema analysis, and scaffolding are delegated to a coordinated set of AI agents.
This is not a solo project that used AI as a productivity tool. It is a different development model. One architect setting direction, agents executing with precision. The 4078-test suite covering every parser edge case was not written by hand. It was specified, reviewed, and shipped through that workflow.
The result is public and auditable. If you want to see what this model produces in practice, start with the repository.
EcoFlow Energy exists because I needed more than a working connection. The reverse engineering, the Protobuf parsers, the reconnect logic, the multi-device architecture. None of that was required to ship something functional. It was required to ship something reliable.
This is how I approach every technical problem. The code is public. Draw your own conclusions.
Buy me a coffee