Support or Ignore reasoning field in response

[yokota]

Hi, support.

I use AI Positron Enterprise with Local LLM.
I had updated the local LLM Ollama recently,
since then, AI Positron fails to catch response and says "Invalid sse format."
It seems to occur by "reasoning" and "reasoning_content" fields in responsed JSON.
Now, I set proxy-server to remove these fields and work with AI Positron.
I hope AI Positron natively supports or ignores these fields.

Regards,

[Cosmin Duna]

Hi,

Thank you for the detailed report.
You're right that the current OpenAI connector (based on chat/completions) does attempt to process reasoning fields, but its handling of them isn't ideal, which is what you're running into with the "Invalid SSE format" error after the Ollama update.

The good news: in the next version of AI Positron, which we're planning to release in the coming few weeks, we're adding support for the Responses API. Once that's available, you'll be able to connect to Ollama through the Responses API endpoint, which should handle reasoning fields (including `reasoning` and `reasoning_content`) properly .

We'll update this thread when the new version is out!

Best regards,
Cosmin

[yokota]

> The good news: in the next version of AI Positron, which we're planning to release in the coming few weeks, we're adding support for the Responses API.

That makes me happy. I'll wait for the release.

Thank you!

[Cosmin Duna]

Hi
I wanted to inform you that Oxygen AI Positron 8.3 has been released and it includes the new OpenAI Response connector type: https://www.oxygenxml.com/doc/ug-positr ... 3c_djv_2hc

Best regards,
Cosmin

[yokota]

Hi,
I updated my AI Positron and set reasoning "none" works now. Thanks!

Still it seems to cause error with the parameters expecting reasoning field; "high", "medium", "auto"...
Is it my settings wrong or the model specific happening?

Oxygen XML Editor is 28.1 latest at Windows11.
AI Positron version is 8.3.0.
Ollama version is 0.31.1.
A model is gemma4:31b-it-qat.

Best regards,

[Radu]

Hello Tatsuro,
Can ypu find the time to enable logging in AI Positron, generate a log file while reproducing the error with the thinking params, then zip and email it to us (support@oxygenxml.com)?
https://www.oxygenxml.com/doc/ug-positr ... oting.html
Regards,
Radu

[yokota]

> Can ypu find the time to enable logging in AI Positron, generate a log file while reproducing the error with the thinking params, then zip and email it to us

The cause might be found at setting to reproduce the problem.
I had used the old settings that labeled now as "OpenAI Chat Completions". In the setting, reasoning other than "none", causes the error.
New setting with OpenAI Responses works fine without the error.

Thanks for great supports.

[Radu]

Hi Tatsuro,
Great, I'm glad!
Regards,
Radu