Purpose

Automatic embedding generation is triggered by SQLAlchemy ORM instance listeners on EmbeddableMixin (after_insert / after_update). Callers no longer need to manually enqueue embedding work after saving an entity. The pipeline passes identity + precomputed searchable text into EmbeddingService, the generator, and the Celery task, so background work does not depend on holding the ORM instance or re-loading it just to get text.

What changed

EmbeddableMixin (mixins.py)

• Still provides searchable_text_changed(), which hashes to_searchable_text() and compares it to Embedding.text_hash rows (first embed, or stale when no row matches the current hash).
• New: @event.listens_for(EmbeddableMixin, "after_insert", propagate=True) and "after_update" handlers build EmbeddingService(session) and call enqueue_embedding(...) with entity_type, entity_id, searchable_text, user_id, and organization_id.

EmbeddingService / generator / task

• enqueue_embedding is keyword-based and passes primitives (plus searchable_text) instead of (entity, current_user).
• EmbeddingGenerator.generate accepts optional searchable_text; when set, it can embed without loading the entity for text.
• generate_embedding_task accepts searchable_text and forwards it to the generator.

Tests

• Updated for the new enqueue_embedding and internal _execute_sync / _enqueue_async signatures; mocks are reset where commit runs and triggers the new listeners.