Onidia is a body without a mind on her own. The model is her brain: you ask her something, your words travel to the model, and the model writes the answer that she speaks or shows you. Everything she does — her replies, her reactions, the way she puts words together — comes from the model you choose.
She can talk to several model providers. Pick the one that fits your setup — each one is a card below.
Google's models via AI Studio. Free tier, no card needed, five minutes to start.
Anthropic's Claude on AWS. Pay per token, no key written into her config.
Local LLM on Raspberry Pi 5 with Hailo AI HAT+. Runs on the NPU, nothing leaves your home.
Access many models through one endpoint. Pick from a catalog of open and commercial models.
OpenRouter is a router, not a single model. You point Onidia at it and choose from its catalog: open models, commercial models, new releases. Good if you want to try different models without creating accounts everywhere.
You still need an OpenRouter API key and a model id from their catalog. The setup is the same shape as the other providers — key in the config, provider set to match.
For day-to-day chatting you do not need the biggest model available.
Both are fast, inexpensive, and more than capable of ordinary chat. Nova Lite on Bedrock or Gemini Flash on Google AI Studio will answer quickly and keep the cost near zero. Save the larger models for when you need deeper reasoning or longer context.