jadzia-core — LLM Gateway Operational

What happened

jadzia-core got its first major architectural decision: a single gateway instead of per-feature API calls scattered across repos.

Design choices:

• Single entry point — lib/llm-gateway.ts handles all outbound LLM calls
• Provider abstraction — Anthropic and Google GenAI behind a common interface
• Env-driven switching — LLM_PROVIDER=anthropic|google toggles without code change
• Retry + fallback — exponential backoff, automatic fallback to secondary provider

Evidence

• All existing Anthropic calls migrated to gateway: 0 direct SDK imports outside lib/
• Google GenAI integration tested: Gemini Pro responses validated against Anthropic baseline
• Provider switch verified end-to-end in staging with LLM_PROVIDER=google

Next step

Add streaming support and token-usage telemetry per provider.