Hot-Swap Any Provider

Picks the Right Brain for the Job · Shipped · workstation, home

Switch between Ollama, LM Studio, Apple Foundation Models, OpenAI, Anthropic, Google, and Azure without restarting.

A headless model picker package that renders provider cards, discovers local models, and lets you assign any discovered model to any tier slot — all without restarting the daemon or losing conversation context.

The @halbert/model-picker package is fully headless: zero I/O, zero hardcoded class names. It defines a ModelPickerTransport interface that the host implements. Endpoint discovery probes each configured provider’s API for available models. The picker supports 8 provider types: Ollama, LM Studio, OpenAI, OpenAI-compatible, Anthropic, Google, Azure OpenAI, and Apple Foundation.

Limits and invariants

The picker never performs fetch calls directly — all I/O goes through the injected transport, so it works identically in tests, Tauri IPC, and REST contexts.

Where this lives

• packages/model-picker/src/useModelPicker.ts
• packages/model-picker/src/types.ts