[Feature Request] Allow for multiple, simultaneous Ollama AI Providers

[c4ch3c4d3]

Currently, Kiln will only allow for a single Ollama provider at a time. While you can specify a custom end point, there's no way to add a second, or third instance on other systems. This would be highly useful for local flows where an end user wishes to run a large model as a judge for evals, while running the final result of a fine tune on a smaller, or different system.

Checks

• Through the current UI, there is no way to specify additional Ollama servers, only a disconnect button.

Describe the solution you'd like
A UI similar to the existing Custom API flow, except for Ollama. This could even be invoked through the existing "hidden" flow that allows a user to first specify a custom endpoint, keeping Kiln simple for those dabbling for the first time.

Describe alternatives you've considered
While Kiln supports Custom API endpoints, and while Ollama supports an OpenAI API endpoint, Kiln is significantly less predictable and reliable when utilizing this alternative method. As an example, during Qwen3-30b-a3b-2507-Instruct works reliably through the standard Ollama endpoint, but exhibits strange or failing behavior when hitting the OpenAI endpoint instead.

[scosman]

Kiln is significantly less predictable and reliable when utilizing this alternative method. As an example, during Qwen3-30b-a3b-2507-Instruct works reliably through the standard Ollama endpoint, but exhibits strange or failing behavior when hitting the OpenAI endpoint instead.

Ah makes sense. The Kiln model library uses knowledge of each platform/model pair to make this works more smoothly.

Try setting JSON Mode = JSON schema, which is what we do on the official Ollama endpoint. That will likely resolve your issue. You can see the settings we use for each provider in the source file ml_model_list.py

Will leave this open to track the request, but no ETA. We have a lot cooking.