What should I use if I need frequent embedding updates and deletes?

If you need frequent updates and deletes of embeddings, the best choice is usually a vector database or search system that supports dynamic indexing well, rather than a static ANN index built for mostly-read workloads.

Good options

• Postgres + pgvector
  Good if your scale is moderate and you want simple update/delete semantics with transactional consistency.
• Milvus
  Designed for vector workloads and handles inserts/updates/deletes better than many “frozen” ANN setups.
• Qdrant
  Strong choice for frequent mutations; built for real-time upserts and deletions.
• Weaviate
  Also supports dynamic updates/deletes and is easy to use at application level.
• OpenSearch / Elasticsearch k-NN
  Useful if you already use search infrastructure and need hybrid text + vector search.

What to avoid

If your workload has constant churn, avoid depending heavily on:

• FAISS indexes in a naive production setup, especially if you need lots of deletes/updates
• static ANN structures that are expensive to rebuild

Rule of thumb

• Small/medium scale, simple ops → Postgres + pgvector
• High-frequency updates/deletes at scale → Qdrant or Milvus
• Need hybrid keyword + vector search → OpenSearch/Elasticsearch

If you tell me your rough scale (number of vectors, update rate, latency target), I can recommend a specific option.