Shared AI Client

MindStore now has a shared AI client at src/server/ai-client.ts.

What It Handles

• loads provider settings from the database
• decrypts sensitive provider keys stored in the database
• resolves the active text-generation provider
• resolves the active transcription provider
• calls OpenAI-compatible, Gemini, and Ollama text models
• calls OpenAI Whisper and Gemini for audio transcription
• streams chat responses for OpenAI-compatible, Gemini, and Ollama providers

Why It Exists

Before this module, AI-heavy plugins copied the same provider logic into each route.

That was risky for convergence because every new plugin port would duplicate:

• provider preference handling
• env fallback behavior
• model defaults
• request formatting
• error handling

Current Users

• Flashcard Maker
• Voice-to-Memory
• Chat

Next Users

• Contradiction Finder
• Blog Draft
• Resume Builder
• Conversation Prep
• Image-to-Memory

Design Rule

Plugin routes and chat routes should not own provider-resolution logic anymore.

They should call:

• shared AI config helpers
• shared AI execution helpers
• plugin-specific business logic in src/server/plugins/ports/*