@onenomad/cortex-llm-core

Contract between pipelines and LLM providers.

Exports:

• LLMProvider — interface every provider package implements
• LLMProviderFactory — (config) => LLMProvider
• LLMRequest, LLMResponse, LLMMessage — the over-the-wire shapes
• TaskPurpose — declarative task labels pipelines use instead of picking models directly (e.g., structural, synthesis, brief, classify)
• LLMRouter — resolves a task + optional model override to a provider, manages a fallback chain, reports health

The router is what pipelines call. They never instantiate providers directly and never pick models — that's all config-driven.