what vector database should I use for embeddings and metadata filtering?

It depends on your scale, latency needs, and how much filtering you need, but here’s the practical shortlist:

Best general-purpose picks

1. Postgres + pgvector

Best if: you want simplicity, strong metadata filtering, and your scale is moderate.

• Great for:
  • embeddings
  • structured metadata filters
  • transactions
  • easy operations if you already use Postgres
• Pros:
  • one database for vectors + metadata
  • SQL filtering is very flexible
  • easy to develop and deploy
• Cons:
  • not ideal for very large-scale ANN workloads
  • performance may lag specialized vector DBs at high volume

Use this if you’re building an MVP, internal tool, or app with up to millions of vectors and need robust filtering.

---

2. Pinecone

Best if: you want managed infrastructure and production-grade vector search with filtering.

• Great for:
  • low operational overhead
  • fast vector search
  • metadata filtering
• Pros:
  • fully managed
  • scales well
  • easy API
• Cons:
  • can be expensive
  • less flexible than SQL-backed approaches
  • vendor lock-in

Use this if you want to move fast and don’t want to manage infrastructure.

---

3. Weaviate

Best if: you want a feature-rich vector database with good metadata support.

• Great for:
  • hybrid search
  • schema-based data
  • filtering
• Pros:
  • strong feature set
  • supports hybrid keyword + vector search
  • open source + managed options
• Cons:
  • more moving parts than Postgres
  • operational complexity if self-hosted

Use this if you need more advanced search features and want an open-source option.

---

4. Qdrant

Best if: metadata filtering is important and you want a strong vector-native system.

• Great for:
  • filtering
  • payload-based metadata
  • fast ANN search
• Pros:
  • excellent filter performance
  • simple API
  • open source and managed offerings
• Cons:
  • not as natural as SQL for complex relational queries
  • smaller ecosystem than Postgres

Use this if you want a dedicated vector DB with good filtering and solid performance.

---

Quick recommendation by use case

• Small/medium app, strong metadata filtering, easiest stack:
  Postgres + pgvector
• Managed production system, minimal ops:
  Pinecone
• Open-source vector DB with strong filtering:
  Qdrant
• Hybrid search + richer vector features:
  Weaviate

---

My default recommendation

If you’re unsure: start with Postgres + pgvector unless you already know you’ll need very high-scale vector search or low-latency managed indexing. It’s usually the best combination of:

• simplicity
• filtering power
• cost
• portability

If you want, I can also recommend one based on your:

• expected number of vectors
• filter complexity
• latency target
• budget
• cloud/on-prem preference