"""Phase 9 — AI-native layer.

A thin, testable foundation for AI-driven features:

* ``providers``     — pluggable LLM / chat backends (echo provider built in).
* ``embeddings``    — embedding backends (deterministic hash backend built in).
* ``vector_store``  — pluggable vector storage (in-memory backend built in).
* ``prompts``       — prompt template registry with safe rendering.
* ``memory``        — short / long-term conversation memory.
* ``context``       — tenant-scoped context builder feeding agents safely.
* ``semantic``      — semantic registry of entities / fields / relations.
* ``actions``       — declarative business operations agents can invoke.
* ``tools``         — agent tools (wrap actions or arbitrary callables).
* ``agents``        — minimal agent runner orchestrating provider + tools.
* ``models``        — AI audit trail (suggestions / actions / approvals).
* ``services``      — high-level façade over the registry + audit trail.
* ``views``/``urls`` — HTTP API.

Everything is offline-by-default — no network calls happen unless a provider
that performs them is explicitly registered.
"""