[Support] NextCloud Assistant Delayed Response Assistance

The Basics

  • [Nextcloud Hub 26 Winter] ((33.0.6) - AIO
  • Host OS: MacOS (Mac mini M4)

Summary of the issue I’m facing:

I’m just starting to get NextCloud Assistant configured in my environment. I have it setup and operational with my local Ollama service. I’m running into an issue with the timing of the AI responses. I read several articles about the NextCloud assistant running on a timer every 5 minutes. I’m running into that bottleneck. I’ve spent most of the day trying different solutions, but have been unable to figure out a way to get NextCloud Assistant to return queries instantly.

I found online reference to running a command similar to this:

php occ background-job:worker -v -t 300 “OC\TaskProcessing\SynchronousBackgroundJob”

If I run that command in Docker Desktop’s console, the NextCloud Assistant will provide response instantly. AI queries previously queued will process immediately as well once that command is executed.

My host computer is MacOS. I tried setting up a crontab job using “crontab -e” so that the service stays alive and restarts every 5 minutes, but could not get it to function.

Example:
*/5 * * * * docker exec --user www-data -d nextcloud-aio-nextcloud php occ background-job:worker -v -t 300 “OC\\TaskProcessing\\SynchronousBackgroundJob” > /dev/null 2>&1

Is anyone aware of any solutions - whether native to NextCloud, in its container, or external like what I’m attempting so that I can have the Assistant respond in real-time? I was hesitant to make any changes to the container at risk of losing said changes during container updates.

Thanks so much in advance for any guidance - Jeff

Have you checked this? How to improve AI task pickup speed in AIO · nextcloud/all-in-one · Discussion #5430 · GitHub

This is how I managed to fix the Assistant pick up speed.
You need to create a script, and configure the crontab to start it after a reboot. Works for me.

But TBH, I definitely gave up trying to use the Assistant at all after many many attempts to maintain the clumsy AI integration stack it relies on. It worked almost well for a while. But it easily gets broken, and is hardly useful for anything.