Files
4d345713fb fix(rag): support 2048-dimensional pgvector embeddings (#3133)
jina-embeddings-v4 (auto-provisioned by LangBot Models in every Cloud
workspace) and doubao embeddings output 2048 dimensions. The shared
pgvector table only enabled 384/512/768/1024/1536/3072, so the first
ingestion into such a knowledge base was rejected with
"Embedding dimension 2048 is not enabled for this deployment".

- Add alembic 0055 widening the enabled-dimension check and creating a
  partial HNSW index over halfvec(2048) (vector HNSW caps at 2000 dims).
- Enable 2048 in the persistence validator, Cloud bootstrap/release
  migration whitelist, pgvector defaults and the config template.
- The rejection message now lists the enabled dimensions.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
2026-11-22 00:06:11 +08:00
..