What is the best option for storing embeddings and doing similarity search?

The “best” option depends on your scale, latency needs, and how much operational complexity you want. In practice, the most common choices are:

Best overall for most teams

PostgreSQL + pgvector

• Great if you already use Postgres
• Simple to operate
• Supports vector similarity search plus normal SQL filtering
• Good for small to medium scale and many production apps

Use this when: you want the simplest reliable solution and your dataset is not enormous.

---

Best for high-performance dedicated vector search

A vector database like Pinecone, Weaviate, Milvus, or Qdrant

• Built specifically for embeddings and similarity search
• Better retrieval performance and scaling than general-purpose databases
• Often easier for large-scale semantic search workloads

Use this when: you need high throughput, low latency, or millions to billions of vectors.

---

Best if you already use a search engine

Elasticsearch / OpenSearch

• Good if you need hybrid search: keyword + vector
• Useful for enterprise search and filtering
• Strong existing ecosystem

Use this when: you want traditional text search and vector search together.

---

Best for analytics-heavy workflows

A data warehouse with vector support Examples: BigQuery, Snowflake, Redshift

• Convenient if embeddings live alongside analytical data
• Not usually the fastest for interactive similarity search
• Better for batch and analytics use cases

---

Quick recommendation

If you’re unsure, choose:

1. Postgres + pgvector for most apps
2. Pinecone/Qdrant/Weaviate/Milvus if you expect significant scale or need specialized vector search
3. OpenSearch/Elasticsearch if hybrid text + vector search is important

If you want, I can recommend the best option based on your exact use case, like:

• number of vectors
• query volume
• filtering needs
• budget
• cloud/on-prem preference