Bluetooth devices are dropped daily (v13.4.0)

Please keep watching the changelog for Release Candidate versions, there might be soon new comming in, as Athom just released “stable”.

I strongly believe this is a power management issue and that the ble power was reduced in 13.4.x update. Hence, the communication range decreased drastically.

I have moved sensors and homey closer, and have improved connection stability now. However, the sensors are not where I want them now, wrong room, so really not a solution for me, only a test.

I have a similar issue, in summary below

Homey Pro (Early 2023), 13.5.3: BLE central connections to a GATT peripheral fail intermittently; same peripheral works every time from a laptop

Summary

A custom Homey app (SDK 3, local, permission homey:wireless:ble) talks to a Bluetooth LE wall panel (Fresh Flow “Fresh Kontroll”, ventilation controller). Since Homey firmware 13.4.0 the connection from the Homey fails often: the panel is missing from scans, connect() fails, or the link drops during service discovery or the first write. The same panel, from a Windows laptop sitting next to it, connects on the first attempt every time. This looks like the BLE regression reported by several users since 13.4.0 (thread above).

Environment

  • Homey Pro (Early 2023), firmware 13.5.3 (also seen on 13.4.x)
  • Homey app: own app, SDK 3, platforms: ["local"], permission homey:wireless:ble
  • Peripheral: BLE panel advertising as “Fresh Kontroll” (address 70:b3:d5:xx:xx:xx), one connection at a time, login by writing 4 bytes to a GATT characteristic, then command and status traffic on a second characteristic
  • Signal: RSSI reported by discover() about -69 to -90 dBm (often rssi=0 and scan=13 ms when an old advertisement is returned from a cache)
  • Wi-Fi: Homey is on a 2.4 GHz network behind an ASUS AiMesh router
  • Other BLE devices on this Homey: [fill in: none / list, before you post]

What the app does (per attempt)

  1. homey.ble.discover() and pick the advertisement by localName or address
  2. advertisement.connect()
  3. peripheral.discoverAllServicesAndCharacteristics()
  4. peripheral.write(service, characteristic, key) (login)
  5. command writes and status reads, then peripheral.disconnect()

The app serialises everything: only one session runs at a time and a failed attempt disconnects before the next one. It retries up to 4 times, each with a fresh scan.

What happens (5 Nov 2026, 10:17 to 13:16)

Typical failures, in the order the steps run:

  • discover() returns 21 to 47 devices but not the panel (found=no) on about half of the scans, although the panel advertises all the time
  • Could not connect to peripheral at connect() (several tries in a row)
  • Peripheral disconnected during service discovery
  • Operation failed with ATT error: 0x0e at the login write, or at the first status read

Excerpt (app log, times local; panel address shortened):

13:02:28 scan try=1 devices=43 found=yes
13:02:36 RETRY try=1/4 step=connect err=Could not connect to peripheral 70b3d5xxxxxx
13:02:38 scan try=2 devices=27 found=yes
13:02:44 RETRY try=2/4 step=connect err=Could not connect to peripheral 70b3d5xxxxxx
13:02:49 scan try=3 devices=26 found=yes
13:02:56 command sent cancel 730ms
13:02:59 SENT-NOSTATUS ... services=4028ms login=1477ms statuserr=Operation failed with ATT error: 0x0e

13:07:38 scan try=2 devices=27 found=yes
13:07:45 RETRY try=2/4 step=connect err=Could not connect to peripheral 70b3d5xxxxxx
13:08:00 scan try=3 devices=44 found=yes
13:08:07 RETRY try=3/4 step=login err=Operation failed with ATT error: 0x0e
13:08:15 scan try=4 devices=28 found=yes
13:08:21 command sent cancel 487ms
13:08:27 OK ... connect=1220ms services=3592ms login=641ms status=6729ms rssi=-90

13:14:43 scan try=1 found=yes
13:14:57 RETRY try=1/4 step=connect err=Could not connect to peripheral 70b3d5xxxxxx
13:15:10 RETRY try=2/4 step=connect err=Could not connect to peripheral 70b3d5xxxxxx
13:15:39 RETRY try=3/4 step=connect err=Could not connect to peripheral 70b3d5xxxxxx
13:16:05 FAILED step=connect ... gave up after 4 tries rssi=-89

Over the afternoon a session usually needed 3 to 4 tries to get through, and some sessions did not get through at all (10:17 to 11:34 and 12:01 to 12:15 failed completely; one session at 11:47 to 11:51 worked).

Control experiment

The same panel was driven from a Windows laptop (Python, bleak), sitting next to the panel, with the same login and the same traffic:

  • 12:27, 12:29, 12:41, 13:01 on 5 Nov: connected on the first attempt every time, login accepted, all reads and writes succeeded, 4 of 4 sessions.
  • So the panel and the application protocol are fine. Only the Homey central is unreliable.

Other observations

  • A Homey reboot gives a window (hours up to a day) in which connections work; an app restart does not help. This matches the other reports in the thread.
  • The signal is weak (-79 to -90 dBm), which probably makes things worse. But the laptop next to the panel works every time, and other users in the thread report the same kind of failures with different BLE devices, so signal alone does not explain the pattern.
  • Because the panel allows one connection at a time, a half-open connection that the Homey does not close cleanly can block the next attempt. The app always calls disconnect() with a timeout before the next try.

Expected behaviour

connect(), service discovery and writes to a BLE peripheral at moderate signal succeed as they did before 13.4.0 (the same hardware worked on earlier firmware for other BLE apps according to the thread).

Request

  1. Confirm whether this is the same regression as in issue #454 and the forum thread, and whether it is in the backlog.
  2. Say whether the BLE changes in 13.4.0 can be rolled back or made configurable, or whether a fix is planned for a coming 13.5.x or 13.6 build.
  3. If there are diagnostics Athom wants (BLE debug logs, a test build), tell me what to collect. I can run a test build on this Homey.
  4. Advice for apps: is there a supported way to keep a central connection open, to tune connection parameters, or to scan with filters that is more reliable than discover()?

The issue with BLE reported in the original post for 13.4 was fixed in 13.5. Everything works fine on my end now.