Milvus is too hard to operate for embeddings

If by “operate” you mean deploy, tune, and maintain Milvus for embedding search, that’s a common complaint. Milvus is powerful, but it can feel heavy if you just want “store vectors and query them.”

What people usually find hard:

• Infrastructure overhead: multiple services, storage, coordination, scaling.
• Operational tuning: index choice, memory sizing, compaction, latency vs recall tradeoffs.
• Production complexity: upgrades, backups, monitoring, and failure handling.
• Overkill for smaller use cases: if you only need a few million vectors, it may be more than you need.

If your goal is simpler vector search, consider:

• pgvector if you already use Postgres and want easier ops.
• Qdrant for a simpler vector DB experience.
• Weaviate if you want an integrated platform with decent ergonomics.
• Pinecone if you want fully managed and don’t mind SaaS.
• FAISS if you want a library, not a database.

A practical rule:

• Need minimal ops → pgvector / managed service
• Need standalone vector DB, simpler than Milvus → Qdrant
• Need large-scale, distributed, high-performance → Milvus can still be worth it

If you want, I can help you:

1. pick a replacement based on your scale and constraints, or
2. simplify your Milvus deployment so it’s easier to run.