@timconsidine Thanks but I had packaged it a while ago as a test but never published it.
AppDev
Private
Cloudron App Packagers
Posts
-
Agentic AI choices -
Agentic AI choices@andreasdueren well done on packaging Hermes so fast !
I still like Windmill, AgentZero and nanoclaw.
Not so much the others, nothing wrong with them, they just don't gel with me.I've started building personal custom apps like dedicated RAG (just RAG and query, nothing else), simple chat, custom-trained models for special purposes. Still exploring whether that delivers better (jury is out currently).
-
Hermes Agent ⚕ -
Agentic AI choices@timconsidine I used nanoclaw but switched to Hermes when Claude closed down the subscription. Thinking about going back to nanoclaw though since they added more auth options besides Claude.
-
Issue pulling custom app from registry on the same cloudron@james good points, thank you.
Honestly … I didn’t check for 9.1.7, lost too much time getting stable operations again, didn’t want to risk further. I will get brave and do it when a process slot appears.
Revoke 9.1.6 : Yes, maybe knee-jerk overkill, but it’s undeniably broken (not Cloudron fault I accept) in a fairly fundamental way for a possibly/probably large number of users. However if 9.1.7 is out and reliable, then users have a path forward. If they know about it.
That’s where the Urgent Advisory Notice is relevant : please check if you’re on kernel xxxx, we have detected incompatibility of 9.1.6 with that, you can fix by 9.1.7 which is available now.
I didn’t choose to deploy 9.1.6, it happened on a reboot and I was left with a broken system. An Advisory would have shortened restoration of full functionality.
-
Issue pulling custom app from registry on the same cloudronFWIW. I have just run into the issue.
Broke connectivity to my cloudron-installed forgejo
Fixed using the patch after losing 4 hours to checking it was not something I did.IMHO :
- EITHER 9.1.6.should be pulled
- OR 9.1.7 should be accelerated
- AND some kind of a clearer higher profile URGENT ADVISORY should be issued for people to check if uname shows 110
I may have missed such an announcement, but I assume this affects quite a number of Cloudron users, so it should be addressed.
Probably a stupid perception issue but backups on patched 9.1.6 seem slower. Working but much slower than before. Will keep an eye on it and try to get some metrics (to rule out user perception).
Seems that rysnc location (hetzner Storage Box) needed remounting. No idea why, but seeks ok now. Watching. -
CloudronVersions.json weirdness@James thank you, will check it out and report back
-
CloudronVersions.json weirdnessAnd again (or so it seems).
Custom App under active development, resulting in frequent additions to CloudronVersions.json and growing file size (~2000 lines).
Update fails
Trimmed old versions content so file is now ~200 lines, just the initial 0.0.1 version and the latest versions.
Update starts working again.Maybe it is some local weirdness (my problem), so would be good to know if others see similar issues when CloudronVersions.json grows large, and if Cloudron team have encountered this.
-
Mail app - FTS indexing fails on mailboxes with large attachments (HTTP timeout)@paradoxbound thanks for investigating . I will fix the timeouts .
-
NetBird - WireGuard based VPNSmall update : although fearful of breaking my production Cloudron, I installed the Netbird client on the box (via command line outside the Cloudron environment - sorry Cloudron team).
Didnetbird up -m <separate VPS running management server> -k <netbird key>.
Worked flawlessly.
On management server, added my Cooudron netbird peer into my private network.
Now apps on Cloudron can reach my laptop, and vice versa, via private network.Specifically a Cloudron app can use Ollama on my laptop running a private model to benefit from Mac Silicon "nearly-gpu-like" performance.
My Cloudron Ollama can't properly run the model because it is CPU only.
Cloudron Ollama running local model on CPU runs at ~6 tokens, or lower.
Mac Ollama private local model achieves ~40 tokens/second.
Not as good as on-demand GPU instance at 80-150 t/s or running an ollama:cloud model also at 80-150 t/s (model dependent).
But 40 t/s is not too shabby and and is viable performance without the cost of an always-on or on-demand GPU.So getting Netbird management server on Cloudron would definitely be a good thing. Shall try to pick that up shortly.
