[Feature Request] local network LLM connection

[SeveranExp]

Please add option to point stack Chan to a locally running model, e.g.: ollama, or other cloud models (OpenAI, Gemini, etc) as per the decision of the user.

[Nodeswitch]

I second this. Having the ability to use locally hosted models, exposing a compatible API to StackChan would be a great feature.

[thezolon]

This is actually something that was promised during the kickstarter.

[SeveranExp]

This is actually something that was promised during the kickstarter.

Not doubting you, but to have it as supporting information, can you share the link? I haven’t found it.

[SeveranExp]

@Forairaaaaa sorry to bug you on this one, but it matters for my intended use case of stack-chan.

[Forairaaaaa]

Hi, the current official firmware uses Xiaozhi for the voice conversation pipeline. Supporting Ollama or other model APIs directly would require changes to how speech recognition, model requests, and speech synthesis are handled.

For an alternative approach, you may want to look at the community project AI_StackChan_Ex . Its custom endpoint guide documents connecting to Ollama, llama.cpp, and LM Studio through an OpenAI-compatible API. Speech recognition and synthesis are configured separately.

This replaces the official firmware, but it may be a useful starting point for your setup.

[Forairaaaaa]

Moving this to Ideas so we can continue discussing the proposal and its scope. Any agreed implementation work can be tracked in a separate issue.