Here's a fun configuration for the self-hosting crowd: a companion that leans on your own machine.
Scowld is bring-your-own-key, and one of the supported LLM providers is Ollama. Point the app at your Ollama setup and your chats run against the model you're hosting yourself. Pair that with native iOS speech recognition for transcription and the core of your voice loop leans on a model you control.
The pieces you'd wire together:
- LLM: Ollama (your own host)
- STT: native iOS speech recognition, no extra key needed
- TTS: optional, via ElevenLabs or OpenAI if you want spoken replies
- Avatar: a VRM character (Aria, Bella, Ciel, or custom) rendered with three-vrm
For anyone who likes keeping their stack close, this is a great starting point. The app provides the companion shell, the VRM rendering, and the voice loop; you provide the brain.
And since Scowld is open source under MIT, you can read exactly how the Ollama provider path is implemented before you trust it with anything.
The code is all there.
Top comments (0)