Embeddings / semantic search

Status: Later
Milestone: After two-week live test
Spec: Journal §4, Processors §7
Package: internal/journal

Goal

Optional semantic search via sqlite-vec virtual table, populated by an embedding processor on write. Keeps vectors in the same SQLite file as events.

Interfaces

Extends search_events to support fuzzy/semantic matching when index exists.

Implementation notes

• Embedding processor module or internal hook on append
• sqlite-vec extension — CGO/SQLite build implications for Pi
• Model choice: local ONNX vs remote API

Acceptance criteria

• [ ] Embeddings stored alongside events
• [ ] search_events returns semantically similar results
• [ ] Degrades gracefully when embeddings disabled

Dependencies

• Blocked by: journal, FTS5 keyword search working first

Open questions

• Embedding model — open-items.md
• AI-derived events non-determinism — spec §7