Issue (Unified SDK): Adopt a unified LLM SDK (Vercel AI SDK)
Summary
We maintain separate adapters for OpenAI and Gemini. Consider a unified SDK to reduce duplication and simplify provider support.
Goals
- Support multiple providers through a single interface
- Reduce adapter maintenance
- Keep config simple
Scope
- Evaluate Vercel AI SDK as primary provider layer
- Map current env config to unified SDK
- Preserve mock provider for local dev
Acceptance criteria
- OpenAI + Gemini work through unified SDK
- Mock provider still works
- Docs updated with new usage/migration notes