Here is the revised support ticket text with the version history clearly stated.
Subject: Assistant not reliably using Ollama backend / Provider routing & LocalAI dependency clarification (Nextcloud 33)
Environment
Nextcloud Version: 33.0.0.16
Previous Working Version: 32.x (same infrastructure)
Database: PostgreSQL (pgsql)
assistant App Version: 3.2.0
integration_openai Version: 4.2.0
Ollama Version: 0.16.1
Ollama Endpoint: http://127.0.0.1:11434/v1
Installed Models (Ollama)
ollama list
NAME ID SIZE
nomic-embed-text:latest 0a109f422b47 274 MB
llama3.1:8b 46e0c10c039e 4.9 GB
Problem Statement
Nextcloud Assistant does not reliably use the configured Ollama backend via integration_openai. Behaviour is inconsistent and difficult to predict.
Observed issues:
• Assistant UI frequently stalls or remains waiting for responses
• Many Assistant interactions produce NO corresponding Ollama logs
• Ollama only logs requests when accessed via curl or OpenWebUI
• Assistant sometimes performs web search instead of local LLM inference
• Disabling LocalAI app causes Assistant failures despite integration_openai + Ollama configuration
This setup previously functioned correctly on Nextcloud version 32 using the same server, Ollama instance, and general configuration approach.
Backend Validation (Working)
Ollama service is running and reachable:
ss -ltnp | grep 11434
LISTEN 127.0.0.1:11434 users:((“ollama”,pid=814,…))
Model API responds correctly:
curl http://127.0.0.1:11434/v1/models
Chat completions succeed:
curl http://127.0.0.1:11434/v1/chat/completions
-H “Content-Type: application/json”
-d ‘{
“model”: “llama3.1:8b”,
“messages”: [
{“role”: “user”, “content”: “hello”}
]
}’
OpenWebUI connected to the same Ollama instance responds within seconds.
Conclusion: Ollama backend is functioning normally.
Nextcloud State (Working)
sudo -u www-data php occ status
installed: true
maintenance: false
needsDbUpgrade: false
Worker Configuration
Dedicated synchronous background workers configured.
Worker script:
/opt/nextcloud-ai-worker/taskprocessing.sh
Contents:
#!/bin/sh
echo “Starting Nextcloud AI Worker $1”
cd /var/www/nextcloud
sudo -u www-data php occ background-job:worker -t 60 ‘OC\TaskProcessing\SynchronousBackgroundJob’
Workers enabled and running:
nextcloud-ai-worker@1.service loaded active running
nextcloud-ai-worker@2.service loaded active running
nextcloud-ai-worker@3.service loaded active running
nextcloud-ai-worker@4.service loaded active running
All four workers are active.
Relevant Configuration
Assistant:
sudo -u www-data php occ config:list assistant
enabled: yes
installed_version: 3.2.0
integration_openai (key values):
provider: custom
url: http://127.0.0.1:11434/v1
default_completion_model_id: llama3.1:8b
chat_endpoint_enabled: 1
ai_integrations:
php occ config:list ai_integrations
{
“apps”: {
“ai_integrations”: []
}
}
Unclear whether this is expected or indicates mis-registration of providers.
Unexpected Behaviour
• Assistant does not consistently trigger Ollama requests
• No Ollama logs for many Assistant prompts
• Assistant appears to fall back to web search without explicit configuration
• LocalAI app state affects Assistant despite using integration_openai
Clarification Requested
-
Is LocalAI required for Assistant even when integration_openai is configured with a custom Ollama endpoint?
-
Should ai_integrations contain provider entries when integration_openai is active?
-
How does Assistant decide between LLM provider vs web search?
-
How can provider routing be validated and debugged?
-
Are there known issues with Assistant + Ollama on Nextcloud 33?
Impact
• Behaviour changed after upgrade from Nextcloud 32 → 33
• Assistant reliability degraded
• Provider selection behaviour unclear
• Local inference cannot be trusted
Request
Please provide:
• Confirmation of correct architecture for Assistant + integration_openai + Ollama
• Recommended diagnostic steps to trace provider routing
• Any known compatibility notes for Ollama backends
Logs, configs, and further diagnostics available on request.