Start with the local path.
Sidekick defaults to local components. A compatible chat model running through Ollama or LM Studio handles reasoning. Whisper provides transcription, Kokoro provides speech and Tesseract reads screen text.
The chat runtime and model are a separate setup. The preview installers include speech and OCR assets. Local models can consume significant memory and storage; model size and your hardware affect speed.
Sidekick
Your conversation and chosen context.
Your chosen model
Ollama or LM Studio, configured separately.
Or bring a supported cloud provider.
Choose a provider available in Sidekick Settings and configure your own account. Provider requests leave your computer. The provider’s privacy rules, model capabilities and prices apply.
Model choice is not feature equivalence.
A model must support the feature you want. Vision is needed to interpret images; tool calling is needed for tool-based tasks. Context limits and output quality also vary. Sidekick’s provider and model selection does not make every model equally capable.
Keep the cost under your control.
There is no paid Retvest subscription for this preview. To avoid cloud model charges, choose a supported local model on hardware you already have. Read a cloud provider’s pricing before using it.