I'm unhappy with my current vector DB because reindexing takes too long

If reindexing is the pain point, the best fix is often not just “a faster vector DB” but a system that supports incremental updates and low-cost rebuilds.

A few practical directions:

1) Use a vector store with fast upserts / incremental indexing

Look for:

• Near real-time upserts
• Background compaction / rebuilding
• Metadata + vector separation
• Sharding / partitioning
• Hybrid search support if you need keyword + vector

Examples people often evaluate:

• Qdrant
• Weaviate
• Pinecone
• Milvus
• OpenSearch / Elasticsearch kNN (if you already use them)

2) Reduce how often you need full reindexing

Common tricks:

• Store embeddings with stable document IDs and update only changed chunks
• Use content hashes to skip unchanged text
• Split large collections into logical partitions
• Keep old index live while building the new one, then swap alias/endpoint
• Re-embed only when the chunk text actually changes, not when parent docs change elsewhere

3) Consider architectural changes

If your data changes frequently:

• Use a hot index for recent updates and a cold index for the bulk corpus
• Periodically merge or compact in the background
• Use a two-stage retrieval pipeline:
  • coarse ANN search
  • rerank a smaller candidate set

4) If the issue is scale, look at operational knobs

Reindexing may be slow because of:

• too many tiny writes
• high-dimensional embeddings
• large payloads attached to vectors
• poor batching
• low HNSW/IVF tuning
• lack of parallelism

You can often improve things with:

• bigger batch sizes
• async ingestion
• tuned index parameters
• separate embedding generation from indexing
• fewer stored fields in the vector DB

5) Sometimes a different retrieval approach is better

If you don’t truly need ANN over a massive corpus, a simpler setup might be enough:

• Postgres + pgvector for moderate scale
• OpenSearch/Elasticsearch for hybrid retrieval
• FAISS if you want a library and can manage the ops yourself

If you want, I can help you narrow this down based on:

1. your data size,
2. how often documents change,
3. current DB, and
4. whether you need hybrid search or just vectors.