Swap the embedding model

• Interface: Embedder (name, min_score, embed(texts)) in apps/ai/app/embeddings.py.
• Today: HashingEmbedder locally (lexical, no model; refused in production) and ProviderEmbedder for any Pydantic AI embedding model, asked for EMBEDDING_DIMENSIONS (1536, apps/ai/app/db/models.py). create_embedder picks one.
• Env: AI_EMBEDDINGS (hashing or a model name such as openai:text-embedding-3-small), AI_MIN_RELEVANCE (cosine floor), and the provider's key (e.g. OPENAI_API_KEY).

Swap

1. Same dimension (1536): set AI_EMBEDDINGS and the key, then re-index every document (vectors from different models aren't comparable; documents keep their text).
2. A different dimension: a migration changing ai.chunk's vector column and its HNSW index (db-change skill), EMBEDDING_DIMENSIONS in apps/ai/app/db/models.py, and a re-index.
3. A model Pydantic AI can't reach: a class with the Embedder shape, chosen in create_embedder.
4. Calibrate AI_MIN_RELEVANCE for the new model with bun run --cwd apps/ai evals.

Tests

apps/ai/tests/test_units.py (embedders), apps/ai/tests/test_service.py (indexing and retrieval against Postgres), apps/ai/tests/test_models_match_db.py after a migration.

Finish

1. The verify skill.
2. Ask the reviewer agent to review the change, and the security-reviewer agent: a new implementation brings its own credentials and sends data somewhere new. If you wrote a migration, the migration-reviewer agent too. The change is in apps/ai, so the python-reviewer agent as well.
3. Update the seam's row in the README's "Scaling path" table if what's "Now" changed.