Ability to pick something else than ollama for llm usage in live dictation

[thibaultmol]

I already have Gemma running in LM Studio, there should be some place in the config to just tell it a (local) url (with optional auth key) that it can pull a list of hosted models from (which would be the local lm studio server in my case)

[rcspam]

Good idea, It was already queued for 1.4. dictee speaks Ollama's own dialect rather than just pointing at its address, so LM Studio needs more than a URL field. The clean answer is the OpenAI-compatible API both of them speak, and the optional auth key you mention fits right in there.

That said, half of what you want is already there today, and I should have said so first. The LLM analysis in the transcription window has its own provider list, and LM Studio is one of the presets, with http://localhost:1234/v1 filled in for you. Settings, LLM Diarization analysis, Manage providers. Pick it, hit Refresh next to the model field and your loaded models show up. Anthropic and any other OpenAI-compatible server work the same way, and there is an API key field.

What is still Ollama-only is the post-processing of your dictation, which talks to a fixed address. That is the part 1.4 opens up.

If you have a preference on how that endpoint setting should look, now is a good time to say it.

Thanks to use dictee