New
- H1232 Ceiling Light Pro gets its ring segments and a separate Main light. The H1232 actually has 17 internal zones behind one masked command — a 16-segment RGBIC ring plus the main panel — but the cloud API only ever exposed 13. This release raises the ring to its full 16 segments and adds a dedicated Main light entity with its own color, brightness, and on/off, all written over the device's native command channel (ptReal) using frames verified against real hardware by the reporter. This needs your Govee account (email/password) logged in to the integration; without it, segments 1-13 still work over the regular API but 14-16 can't be written, and ptReal-only entities now fall back to the regular API automatically so a native-command hiccup doesn't block the segments that don't need it. Scenes still only show on the physical panel. (#223) — if you're on an H1232, please update, test the ring and Main light, and let us know whether the old template-light workaround can be retired. The H60A6 and H1252 are intentionally not included yet since nobody has verified their frames; owners of those models, please hold off until that happens.
- H5059 water sensor: separate Upper probe and Lower probe moisture entities. Previously only a single aggregate "wet/dry" state was exposed. Using a reporter-supplied byte map, the integration now decodes the upper and lower probes independently, so you can tell which probe tripped; the existing combined sensor is unchanged. (#224) — this is unverified against Govee's own cloud event push (as opposed to the device's own status frame), so if you're on an H5059, please download diagnostics right after a trip so a real probe-trip event can be checked.
- Diagnostics now show how each device's segment count was derived. Some SKUs report a RGBIC segment count from the cloud API that doesn't match the real device, requiring a debug-log hunt to track down. The diagnostics download now includes, per device, the raw API count, any size-based clamp, any SKU-specific override, and the final count actually used — useful context if you ever need to report a wrong segment count. (#160, #208, #223)
- MQTT diagnostics record every inbound message. To help diagnose accounts where real-time updates stop arriving, the diagnostics download now includes every inbound MQTT message along with when the last one arrived. (#222)
Fixes
- H605B DreamView "off" no longer leaves the backlight stuck in video mode. Turning off screen sync on the H605B TV Backlight sent the off command and Govee accepted it, but the light itself stayed in video mode instead of returning to a normal color — the same issue previously fixed for the H66A0 (#213). Turning DreamView off now restores the light's last color; use the light entity itself if you want the backlight fully off. (#220) — please confirm this matches what you see in the app after updating.
- H7124/H7129/H7126 air purifiers work again after a previous update broke their Mode select and fan speeds. These purifiers use a
gearModespeed setting instead of thepurifierModecapability the integration assumed, so every mode-select command was rejected by Govee. The Mode select now sends gear-mode speeds correctly for these models; the H6006-style purifiers and the H7121 (#201) are unaffected. (#221) — if you're on one of these models, please confirm both the Mode select and the fan entity's speed presets work after updating. - Connection-mode sensor no longer reports MQTT as active from a stale, previous session. The sensor now only considers the connection "MQTT" when a frame has actually arrived during the current session, fixing a case where it kept showing MQTT connectivity that had actually stopped. (#222)
- Lights on a healthy local network stay available during a cloud outage. Previously, if every cloud state poll failed, the whole-device light would go unavailable even when it was still reachable and responsive over the local network. It now stays available as long as the local (LAN) connection is healthy; other entities that depend on the cloud are unaffected. Thanks @drewcotten for the fix (#226) — the author notes they haven't yet tested this against a real WAN outage on their own hardware, so if you hit this scenario, please report back on the issue with what you see.
- H5103 WiFi thermo-hygrometer reads temperature correctly in Auto unit mode. The H5103 reports its temperature in Fahrenheit internally even though the reading is tagged Celsius, the same quirk as the H5053/H5171 (#173); with "Auto" unit detection, this showed as a wildly wrong value (e.g. 71.2° instead of the correct 21.8°C). The H5103 is now recognized so Auto mode converts it correctly out of the box, without needing the Fahrenheit override workaround from #85. Thanks @usednick for the fix (#225).