Nextcloud Assistant + OpenAI: how to correctly configure and verify the model being used?

[vawaver]

Nextcloud Assistant + OpenAI: how to correctly configure and verify the model being used?

Hi,

I am trying to configure Nextcloud Assistant with the official OpenAI integration, and I would like to ask for advice from people who already use this setup in practice.

My setup:

• Nextcloud: 33.0.3
• Nextcloud AIO: 13.1.0 beta
• OpenAI API account is configured in Nextcloud
• The API key is accepted and the OpenAI model list is visible in the administration settings

In the OpenAI administration settings I can see several available models, including for example:

gpt-5.5 (system)
gpt-5.5-2026-04-23 (system)
gpt-5.5-pro (system)
gpt-5.5-pro-2026-04-23 (system)
chat-latest (system)

I selected:

gpt-5.5 (system)

However, when I ask Nextcloud Assistant something like:

I want to know what model you are.

the answer still says something like:

I am ChatGPT, an AI language model developed by OpenAI, based on the GPT-4 architecture.

I understand that asking the model directly is probably not a reliable way to verify the actual backend model ID. The model may not know what exact API model Nextcloud is using internally.

So my question is more general:

What is the correct way to verify which OpenAI model Nextcloud Assistant is actually using?

I would like to understand:

1. Is selecting the model in the OpenAI integration settings enough for Nextcloud Assistant to use it?

2. Is there another place in Nextcloud where the active AI provider or text generation model must also be selected?

3. Does Nextcloud Assistant always use the selected default model from the OpenAI integration, or can some Assistant features use a different provider/model internally?

4. What is the best way to confirm the real model being used?

   • OpenAI dashboard / usage logs?
   • Nextcloud logs?
   • OCC command?
   • Browser developer tools?
   • Some internal task processing configuration?

5. Has anyone here already configured Nextcloud Assistant with the newer OpenAI models and verified that the selected model is really being used?

The main goal is not only to make the Assistant work, but to understand how Nextcloud decides which OpenAI model is used and how this can be verified reliably.

[edward]

The model chosen in the admin settings is only the default model for all tasks where OpenAI is chosen as the provider. The model to actually use can be changed within the Assistant under the advanced options:

Is that what you are looking for, or are you referring to something else?

[vawaver]

Glad for your answer, but I think this explains the model selection for Work with text.

My question is specifically about Chat with AI.

In Chat with AI, I do not see any model selector, so I am trying to find out how to verify which OpenAI model is actually being used there.

Is there a log entry, config value, or occ command where I can check the real model/provider used by Chat with AI?

For example, can this be verified through:

• nextcloud.log
• an occ config:* command
• integration_openai app config
• assistant / task processing config
• OpenAI request logs

I understand that asking the assistant “what model are you?” is not reliable. I am looking for a technical way to verify it from logs or configuration.

I am not a developer, just a Nextcloud AIO user/admin trying to confirm what model is actually used by Chat with AI.

[vawaver]

Small follow-up after checking this directly in my Nextcloud AIO setup.

My current setup:

Nextcloud: 33.0.3
Nextcloud AIO: 13.1.0 beta
Assistant: 3.4.3
integration_openai: 4.5.1

I wanted to verify specifically which model is used by Assistant → Chat with AI, because I do not see a separate model selector there like in Work with text.

First I checked the configured default completion model in the OpenAI integration:

sudo docker exec --user www-data --workdir /var/www/html nextcloud-aio-nextcloud php occ config:app:get integration_openai default_completion_model_id

In my case the output was:

gpt-5.5

Then I checked the installed integration_openai code directly inside the AIO container:

sudo docker exec --workdir /var/www/html nextcloud-aio-nextcloud sh -lc '
APP_DIR=$(find /var/www/html/custom_apps /var/www/html/apps -maxdepth 2 -type d -name integration_openai 2>/dev/null | head -n1)

echo "APP_DIR=$APP_DIR"
echo

grep -R "getAdminDefaultCompletionModelId\|createChatCompletion\|default_completion_model_id\|DEFAULT_COMPLETION_MODEL_ID" -n "$APP_DIR/lib" | sed -n "1,200p"
'

The relevant result was:

/var/www/html/custom_apps/integration_openai/lib/TaskProcessing/TextToTextChatProvider.php:93:
$adminModel = $this->openAiSettingsService->getAdminDefaultCompletionModelId();

/var/www/html/custom_apps/integration_openai/lib/TaskProcessing/TextToTextChatProvider.php:121:
$completion = $this->openAiAPIService->createChatCompletion($userId, $adminModel, $userPrompt, $systemPrompt, $history, 1, $maxTokens);

So for Chat with AI, the installed code path seems to use:

integration_openai/default_completion_model_id

In my case this means:

Chat with AI → integration_openai/default_completion_model_id → gpt-5.5

There is also a fallback value in the app code:

Application.php: public const DEFAULT_COMPLETION_MODEL_ID = 'gpt-4.1-mini';

But this should only matter if no default completion model is configured. Since my default_completion_model_id is explicitly set to gpt-5.5, the fallback should not be used here.

So my current conclusion is:

Assistant → Chat with AI does not seem to have its own separate visible model selector.
It appears to use the default completion model configured in the OpenAI integration.

For a final low-level confirmation, the remaining option would be to temporarily capture the outgoing OpenAI API request and check the actual JSON body for:

"model": "gpt-5.5"

But from the local AIO configuration and installed code path, this already gives a clear technical indication of which model Chat with AI should be using.

[janepie]

It’s the one configured for text generation in the OpenAI integration settings:

I guess we could add a note there that it’s the Chat one

[Daphne]

Chat with AI indeed does not have its own model selector, it is selected by the administrator in the way @janepie described in the post above me.