If any of those backups has a valid /config folder you can access, you can drop it over the top of a new install (get all the hidden folders and subtree) and it should get your base HA at least. Add… Er Apps are separate…
Maybe?
I started an issue here - 2026.9 Update Makes system unstable - Large Memory usage spike · Issue #181120 · home-assistant/core · GitHub
I have managed to get stable on a totally stripped down system then manually trigger it by enabling a variety of integrations.
There was some sort of issue with CPU definition in Proxmox VMs raised during beta testing. I don’t have access to Discord right now to check what it was.
EDIT: Try changing the VM CPU Type to x86-64-v2-AES or host.
I was running host but for science i also tried with x86-64-v2-AES and had the same problem
My new VM, fresh install on proxmox, crashed during the restore, didn’t see much in the way of logs from the proxmox console.
That is set to x86-64-v2-AES, so yeah, don’t think the restore should take 2 hours
The old VM before i bricked it was set to x86-64-v2-AES and i could barely connect before it restarted.
Update:
I think i’m back up and running, sort of, back on 2026.8.3, need to sort out a few things, but working, will see how i go.
For visibility you should also post about this in the release notes topic. 2026.9: There's room on this bus
same issue, beelink, HAOS, can ssh to system, but can’t hit UI
edit: seems to come up in safe-mode
edit 2: crashed in safe-mode
After a Fresh VM install and restore from my backup (core config only) I attempted to update again, It again has the same issue upon completion.
Same issue here with cycling reconnection. RPi 5, HAOS, can not ssh in.
Rolled back to 08.03. No local UI connection. But UI does connect via nabu casa. Also, am using vanity domain.
Same on docker running on RPI4.
Constant crashing making system unreachable.
I can confirm this issue on Home Assistant OS running as a Proxmox VM.
Environment
Home Assistant OS: 18.2
Affected Home Assistant Core version: 2026.9.0
Last working Core version: 2026.8.3
Supervisor: 2026.08.0
Hypervisor: Proxmox VE 8.2.7
VM resources: 2 vCPUs and 6 GiB RAM
CPU type: host
Symptoms
Immediately after upgrading from Core 2026.8.3 to 2026.9.0, Core started becoming unavailable and restarting repeatedly.
The VM itself did not reboot: its Proxmox uptime remained continuous. Supervisor and the Observer on port 4357 remained connected and healthy, while Core on port 8123 became unavailable.
The HAOS console showed repeated OOM killer events affecting both Music Assistant (mass) and Home Assistant Core (python3). One of the Core events was:
text Out of memory: Killed process 3428352 (python3) total-vm:6845188kB anon-rss:4079676kB file-rss:5220kB shmem-rss:0kB UID:0 pgtables:10008kB oom_score_adj:-300
I initially stopped the Music Assistant add-on to rule it out. This temporarily freed resources, but Core was killed by the OOM killer again after being started, so Music Assistant was not the sole cause.
The Core CLI command itself completed successfully:
text ha core start Processing… Done. Command completed successfully.
However, the python3 Core process was killed by the kernel shortly afterwards. The watchdog then attempted to restart it, causing the loop.
Resolution
I resolved the issue by restoring a complete Proxmox VM backup created immediately before upgrading to Core 2026.9.0.
The backup restored the entire Home Assistant OS VM to its previous consistent state, including:
Home Assistant Core 2026.8.3
Supervisor and HAOS configuration
Home Assistant database and integrations
Add-ons
Network identity and USB passthrough
After restoring the backup, the system returned to Core 2026.8.3 and became stable again. Home Assistant, Zigbee, Music Assistant, and the other services are all operational.
In my case, stopping Music Assistant alone did not resolve the problem. Restoring the full pre-update Proxmox VM backup was the successful recovery method.
I have this problem specifically with the Vivaldi web browser when using the cloud address. It works correctly with the Firefox browser and the cloud address, and I don’t have a problem with using the local IP address of my home assistant instance.
I am running HAOS on an x86 HP thin client.
I was slightly apprehensive about updating my bare metal HAOS install, but I decided to anyway because why not 
I have zero issues on my 16GB M710q - update went smoothly, and my RAM usage has stayed exactly the same before and after the update:
total used free shared buff/cache available
Mem: 15Gi 5.6Gi 215Mi 197Mi 10Gi 9.9Gi
Swap: 4.0Gi 2.0Gi 2.0Gi
So I wonder if this is Proxmox specific somehow?
EDIT: maybe not, see comment below
Problem seems like it could be related to a bloated core.device_registry file.
possible fix being worked on here: Resolve colliding orphans through the orphan indexes by joostlek · Pull Request #181201 · home-assistant/core · GitHub
Also prevention on file bloating being worked on here: Purge unreachable deleted devices on device registry load by emontnemery · Pull Request #181207 · home-assistant/core · GitHub
It seems that something misbehaved in the past and put over 700k deleted devices in my registry file.
People more clever than me are confirming and working on it.
Good to hear, i’m now back on 2026.8.3, although i seem to be getting out of memory issues again.
Assigned a little more RAM to the VM(from 8GB to 10GB) but thats about my limit i think.
I had an integration go nutso a couple months ago and create 18000 entites, which caused plenty of memory issues, but a restore to an older version resolved that, i have 1500 or so entities, and 230 devices i think in total currently, so nothing too crazy.
Hoping adding Thread hasn’t added extra complications, it was complicated enough to get working in the first place 
Hi new to this , l have been using HA on greens since 25. l have 6 greens. Not much issues until today. I updated 3 units but shortly after l got reports from users that the remote via Nuba Casa and Tailscale locked them out and rebooting via system it allowed them in but kept dropping connections .. local was fine . We restored to 26.8.3 and seems stable for now .. but why no word from open foundation.. i saw one post from “OF” saying simular issue during beta week, but resolved, its not resolved as we experienced today
Hi, keep an eye for the minor update that most likely will be released tomorrow.
Anybody needing to get their system working again can downgrade by issuing the command:
core update --version 2026.8.2