Issue with Nextcloud Assistant AI Chat - No Response on Questions

, ,

Hello,

I am using Nextcloud AIO version 12.5.0 with Nextcloud version 32.0.4 and Nextcloud Assistant version 2.12.0, integrated via a paid OpenAI API. However, I am encountering an issue with the Nextcloud Assistant Chat module. Whenever I input a question into the AI Chat, I do not receive any response. The system seems to be processing the question (as indicated by the “working” status), but no answer is returned.

Interestingly, other modules such as Text Processing, Translation, Emoticon Generation, and Image Generation all work without issues, providing responses within seconds. Similarly, the Nextcloud Assistant Talk Bot is functioning correctly.

A few versions ago, everything was working fine, but now, since today, I can no longer get a response from the AI Chat.

I have already tried recreating all the containers, but the issue remains.

Has anyone encountered something similar, or can anyone suggest a fix for this?

Hello, I am having the same issue, does anyone found a solution for this ? it takes around 3-4mins for an answer, and when 2 or more users are using the AI chat at the same time none will work or have an answer.

I did some additional testing today and may have found something interesting.

I tested on the following setup:

  • Nextcloud AIO: 12.7.0 (stable)

  • Nextcloud: 32.0.6 Autumn

  • Nextcloud Assistant: 2.13.0

  • Nextcloud Context Agent: 2.3.0

After trying several different combinations, I discovered that Nextcloud Assistant Chat started working only after I disabled (not removed) the Nextcloud Context Agent.

Steps that worked for me:

  1. Disable Nextcloud Context Agent (the Docker container created by it shows the status stopped in Portainer).

  2. Test Nextcloud Assistant → AI Chat started working immediately.

  3. Re-enable Nextcloud Context Agent again.

  4. After re-enabling it, Nextcloud Assistant continued working normally.

So it seems the Nextcloud Context Agent may get stuck in some inconsistent state, and toggling it off/on resolves the issue.

I would appreciate it if someone else could confirm whether this workaround also fixes the problem on their setup.

I want to add a more detailed follow-up to this topic, because after more testing I think I isolated the trigger more clearly:

Current environment:

  • Nextcloud AIO 12.9.2 Beta
  • Nextcloud 32.0.8
  • Nextcloud Assistant 2.13.0
  • Assistant Talk Bot 3.1.3
  • Nextcloud Context Agent 2.5.1
  • Nextcloud Assistant Context Chat 5.3.1

Provider:

  • OpenAI

What I am seeing right now is very consistent:

  • if Nextcloud Assistant Context Chat is disabled
  • and Nextcloud Context Agent is disabled

then both of these work normally again:

  • Nextcloud Assistant
  • Assistant Talk Bot

As soon as I enable either one of these apps again:

  • Nextcloud Assistant Context Chat
  • or Nextcloud Context Agent

then the normal AI responses stop working again.

In other words, from what I can currently reproduce, it is enough to enable one of those two apps to break the behavior of:

  • the regular Nextcloud Assistant
  • and the Assistant Talk Bot

That is why I no longer think this is just a simple UI problem inside one specific app.

At this point my working theory is that there is some shared layer underneath all of this that gets into a bad state once Context Chat or Context Agent becomes part of the chain.

From my testing, this looks less like:

  • a general OpenAI connectivity problem
  • or a problem limited only to the Assistant frontend

and more like something affecting a shared AI pipeline, for example:

  • the common text generation layer
  • the task processing / background job layer
  • or the internal handoff between Assistant / Talk Bot and the apps that extend the AI stack

The reason I suspect that is simple:

  • with Context Chat and Context Agent disabled, Assistant and Talk Bot respond normally
  • with either of them enabled, both stop responding

So in my case, enabling one of these “extra” AI-related apps seems to poison or block the shared path that the other Assistant features rely on.

I do not want to overstate this as a confirmed root cause yet, because I do not have a clean code-level explanation at the moment. But from the outside, that is exactly how it behaves.

I have already been debugging this from multiple angles, including background jobs and the general AI response path, and this is currently the clearest pattern I can reproduce.

That is why I would like to ask others with a similar setup to test this and compare results.

I would especially be interested in feedback from users who run something close to this:

  • Nextcloud AIO
  • OpenAI integration
  • Nextcloud Assistant
  • Assistant Bot in Talk
  • Nextcloud Assistant Context Chat
  • Nextcloud Context Agent

Questions:

  1. Can anyone reproduce that enabling Context Chat breaks normal Assistant / Talk Bot responses?
  2. Can anyone reproduce the same behavior with Context Agent?
  3. If you use OpenAI as provider, do Assistant and Talk Bot still work normally with both of those apps enabled?
  4. Is this reproducible only on AIO / Beta, or also elsewhere?

At this point I am mainly trying to determine whether:

  • this is a broader regression
  • or whether it is something specific to my environment

If someone can confirm or deny this with a similar OpenAI-based setup, that would help a lot.

Right now, in my environment, I am effectively stuck with a choice between:

  • using Nextcloud Assistant + Assistant Bot in Talk
  • or using Nextcloud Assistant Context Chat

because at the moment I cannot get both of those worlds to work together at the same time.

Hi, facing the same issue here with Llama.cpp backend.

Takes 3-4 minutes until the request is sent to the backend; no streaming answer, no message in the logs.

On top of it:

  • Any time I open the assistant setting page, my api key resets to DummyApiKey
  • The completion path change to /v1/chat/completions (llama.cpp expects v1/completions) while the UI shows " completion "

Do I need to mention that this was some very frustrating last 24 hours ? To the point I was starting to think someone tries to sabotage any working local LLM implementation…

I just found your post and will try to follow your instruction to debug that mess and will post my feedback.