Ensure we properly support structured outputs

[dkotter]

Description of the Change

Ollama supports models that support structured outputs but we weren't properly parsing those ourselves. This means any API request that requested a structured output wouldn't work properly.

How to test the Change

1. Pull this PR down and set up the Ollama provider
2. Install the AI plugin and turn on Content Classification and Review Notes (both use structured outputs)
3. Test both and ensure they work, noting that depending on the model you're using, you may run into timeouts on Review Notes

Changelog Entry

Changed - Increase the standard timeout to be 60 seconds
Fixed - Properly parse structured outputs

Credits

Props @dkotter

Checklist:

• I agree to follow this project's Code of Conduct.
• I have updated the documentation accordingly.
• I have added Critical Flows, Test Cases, and/or End-to-End Tests to cover my change.
• All new and existing tests pass.

[jeffpaul]

I'm running 7.0-RC2-62242 and among other models have qwen3.6:latest available, but keep getting this error:

Network error occurred while sending request to http://localhost:11434/v1/chat/completions: cURL error 28: Operation timed out after 30001 milliseconds with 0 bytes received

[dkotter]

I'm running 7.0-RC2-62242 and among other models have qwen3.6:latest available, but keep getting this error:

Network error occurred while sending request to http://localhost:11434/v1/chat/completions: cURL error 28: Operation timed out after 30001 milliseconds with 0 bytes received

I had run into timeouts a few times in testing though not consistently. That 30 second timeout comes upstream from the AI Client but just pushed up a fix that changes that default to 60 seconds when using this provider. That seemed high enough in my testing to never hit timeouts but let me know if you're still having issues.

[jeffpaul]

Not surprisingly, now getting:

Network error occurred while sending request to http://localhost:11434/v1/chat/completions: cURL error 28: Operation timed out after 60002 milliseconds with 0 bytes received

I've got a whole bunch of older models locally but also some newer, larger ones. Perhaps I'm getting routed to a poorly performing model here?

[dkotter]

I've got a whole bunch of older models locally but also some newer, larger ones. Perhaps I'm getting routed to a poorly performing model here?

Hmm.. maybe. The issue we have right now is there's no way to easily check which model was used in the request.

If you can, can you install this version of the AI plugin (this is from PR 437):

pr-437-f588fc861783a4ec9306ca070a51a25e40c6dfd4.zip

Then go to the AI settings page and activate the AI Request Logging experiment. This should then show a page under Tools > AI Request Logs. Now try running Content Classification or Review Notes and then refresh this log page. It should show you what model was used which I can then try and replicate on my end:

[jeffpaul]

---

[dkotter]

Interesting. I'm using the same model but not running into this. Is it consistent across all features? Or just the review notes experiment?

[jeffpaul]

Updated Ollama to 0.21.1 and classification and review notes both work, let's go!