---
title: "KURE embedding pilot shadow index implemented"
from: codex
to: ["claude-code", "jadong"]
project: "maester-agent"
date: 2026-06-20T12:06:16.717657+09:00
status: sent
type: result
---

## Result

- Branch: `wip/kure-embedding-pilot`
- Docs commit: `4946665 docs(context): preserve vector DB embedding research decision`
- Code commit: `d13c556 feat(vectorize): add KURE embedding pilot shadow index`
- Added KURE provider config defaults: `nlpai-lab/KURE-v1`, `1024`, profile `kure`.
- Added Neo4j shadow property/index support: `embedding_kure_1024`, `maester_chunk_embedding_kure_1024`.
- Kept KURE metadata separate as `embedding_kure_model`, `embedding_kure_dim`, etc.
- Added backfill CLI overrides for provider/model/dim/endpoint.

## Verification

- `py_compile` passed for vector/graph/scripts/tools files.
- `scripts/test_vectorize.py` passed.
- KURE dry-run backfill passed for 3 docs.
- Neo4j query/index probe passed with nonzero vector.

## Remaining

Actual KURE vector generation needs either `MAESTER_KURE_ENDPOINT` pointing to an OpenAI-compatible embedding server, or local optional `sentence-transformers`/torch installation. Current maester venv does not yet include sentence-transformers/torch.