Azure AI API Integration Error – "Invalid request"

awild · March 30, 2026, 1:01pm

Infos:

Used Zammad version: 7.0.0-1773148656.3c95b7da.centos9
Used Zammad installation type: package
Operating system: centos9
Browser + version: MS Edge & Google Chrome

Expected behavior:

The Azure AI provider should be successfully connected to Zammad after entering valid API credentials (token + URL),
allowing AI-powered features (e.g., chat completions) to function.

Actual behavior:

When saving the Azure AI configuration under Settings → AI → Provider, the following error appears:
“AI provider is not accessible: Invalid request – please check your input”
The configuration is not saved, and no further details are provided in the UI or logs (if checked).

Steps to reproduce the behavior:

Navigate to Settings → AI → Provider.
Select Azure AI from the dropdown.
Token: [Valid Azure AI API key, confirmed working via direct API calls]
URL (Completions): https://[region].openai.azure.com/openai/deployments/[deployment-name]/chat/completions?api-version=2023-05-15 (format matches Azure’s documentation).
URL (Completions): https://xxxxxxx-swedencentral.openai.azure.com/openai/deployments/GPT5-Zammad-Test/chat/completions?api-version=2025-03-01-preview

Leave OCR disabled (not required).
Click Save.

Error: AI provider is not accessible: Invalid request – please check your input.

dominikklein · March 30, 2026, 1:30pm

Did you check the HttpLog (Feedback & Logs tab) for additional information?

The URL does not look that bad, but maybe it’s something else. Compared to our testing, it’s a “thinking” LLM, which is used; maybe this is causing problems.

I can also try to create a later or tomorrow, some GPT-5 url, and check if it’s working.

awild · March 30, 2026, 1:50pm

Mistral, for example, works perfectly.
Azure AI causes problems.

dominikklein · March 30, 2026, 2:07pm

I mean not AI-Error-Log. It was about the HttpLog (“Recent Logs”) which should be directly visible on this tab.

awild · March 30, 2026, 2:30pm

No log entrys visible for azure:

dominikklein · March 30, 2026, 8:53pm

Ok, normally errors on provider configuration should also be logged. Maybe we missed something here

awild · March 31, 2026, 11:48am

OK, I am waiting for your feedback on how we can proceed. Thank you

fliebe92 · March 31, 2026, 12:30pm

Hi @awild. Do you have any logs in general available from your Zammad instance? production.log for example.

awild · March 31, 2026, 12:57pm

This error is contained in the log file:

I, [2026-03-31T14:51:51.900423#1032335-84192] INFO – : Parameters: {“name” => “ai_provider_config”, “state_current” => {“value” => {“provider” => “azure”, “token” => “[FILTERED]”, “url_completions” => “https://xxxxxx-swedencentral.cognitiveservices.azure.com/openai/deployments/GPT5-Zammad-Test/chat/completions?api-version=2025-01-01-preview”, “url_ocr” => “”, “ocr_active” => false}}, “id” => “284”}
E, [2026-03-31T14:51:52.482667#1032335-84192] ERROR – : Validation failed: AI provider is not accessible: Invalid request - please check your input (ActiveRecord::RecordInvalid)

dominikklein · March 31, 2026, 8:55pm

We checked the Azure provider implementation and know the root cause.

I will create a bug report tomorrow.

dominikklein · April 1, 2026, 1:33pm

github.com/zammad/zammad

Missing temperature support by some models for some providers

opened 01:20PM - 01 Apr 26 UTC

dvuckovic

verified integration AI

### Used Zammad Version 7.0 ### Environment - Installation method: any - Oper…ating system: any - Database + version: any - Elasticsearch version: any - Browser + version: any ### Actual behaviour If one of [the "reasoning" OpenAI models](https://developers.openai.com/api/docs/guides/reasoning) is configured for an AI provider (i.e. `model_id.match?(/^(o\d|gpt-5)/)`), it can raise the following error during AI feature usage: ```json { "error": { "message": "Unsupported value: 'temperature' does not support 0.1 with this model. Only the default (1) value is supported.", "type": "invalid_request_error", "param": "temperature", "code": "unsupported_value" } } ``` ### Expected behaviour Zammad should not send an unsupported parameter value to the AI provider in case we can assume the model will reject the request. ### Steps to reproduce the behaviour 1. Configure a "Custom (OpenAI Compatible)" AI provider, that uses one of the "reasoning" models (`gpt-5.4` is a good choice). 2. Note that the configuration is saved (after a successful "ping" request). 3. Try to use an AI feature with a temperature (e.g. Ticket Summary). 4. Note the aforementioned error in the logs. Although a similar issue was already addressed for the "OpenAI" provider in #5759, with the new models available under other providers, we are facing a similar problem. There doesn't seem to be a "model capability" API we can use to determine if the model indeed supports a custom temperature. Consider implementing a simple "check" which is executed immediately after a successful "ping". In case it raises an error, remember that the temperature should not be used with this model. ### PoC <details> <summary>Failed "check"</summary> ``` $ curl "https://openai-compatible-domain.tld/v1/chat/completions" \ -H 'Content-Type: application/json' \ -H "Authorization: Bearer $OPEN_AI_TOKEN" \ -X POST \ --data '{ "model": "gpt-5", "messages": [{ "role": "user", "content": "whats your name?" }], "stream": false, "store": false, "temperature": 0.5 }' \ -w "\n%{http_code}\n" { "error": { "message": "Unsupported value: 'temperature' does not support 0.5 with this model. Only the default (1) value is supported.", "type": "invalid_request_error", "param": "temperature", "code": "unsupported_value" } } 400 ``` </details> <details> <summary>Successful "check"</summary> ``` $ curl "https://openai-compatible-domain.tld/v1/chat/completions" \ -H 'Content-Type: application/json' \ -H "Authorization: Bearer $OPEN_AI_TOKEN" \ -X POST \ --data '{ "model": "gpt-4", "messages": [{ "role": "user", "content": "whats your name?" }], "stream": false, "store": false, "temperature": 0.5 }' \ -w "\n%{http_code}\n" { "id": "chatcmpl-DPlAgtku2MGmzO2ltx4uBAs2hwRT3", "object": "chat.completion", "created": 1775031898, "model": "gpt-4-0613", ... } 200 ``` </details> ### Support Ticket Ticket#10219325, Ticket#10219962 ### I'm sure this is a bug and no feature request or a general question. yes

awild · April 13, 2026, 8:28am

New info, it works with GPT-4o:

dominikklein · April 14, 2026, 7:32am

Yes, we know, because it has no problems with the temperature We have also the fix in the pipeline, should be available during the week, let’s see how it’s going.