Vector databases are the backbone of any RAG system. They store document embeddings and enable fast similarity search — finding the most relevant chunks of text for a given query. But with dozens of options available, choosing the right one matters.

What Makes Vector Databases Different

Traditional databases index structured data for exact matches. Vector databases index high-dimensional numerical representations (embeddings) and find approximate nearest neighbours. This enables semantic search — finding documents that are conceptually similar to a query, even if they don't share exact keywords.

Key Options Compared

Azure AI Search — Microsoft's managed search service with built-in vector search. Great for Australian enterprises already on Azure. Supports hybrid search (vector + keyword) out of the box. Managed service means less operational overhead.

Pinecone — Purpose-built managed vector database. Excellent developer experience and performance. Serverless pricing model works well for variable workloads. No Australian region yet, so latency considerations apply.

Weaviate — Open-source vector database with a managed cloud option. Flexible schema, built-in vectorisation modules, and strong hybrid search. Can be self-hosted for data sovereignty requirements.

Qdrant — High-performance open-source option written in Rust. Excellent for latency-sensitive applications. Cloud and self-hosted options available.

pgvector — PostgreSQL extension for vector similarity search. If you're already running Postgres, this adds vector capabilities without a new system. Good for smaller datasets; performance degrades at scale compared to purpose-built solutions.

Decision Framework

Consider these factors:

1. **Scale** — How many vectors? Millions require purpose-built solutions. Thousands can work with pgvector.
2. **Latency requirements** — Real-time chat needs sub-100ms retrieval. Batch processing is more forgiving.
3. **Data sovereignty** — Australian data residency requirements may limit cloud options.
4. **Operational complexity** — Managed services cost more but require less engineering time.
5. **Hybrid search** — Combining vector and keyword search typically improves RAG accuracy by 10-20%.

Our Recommendation

For most Australian enterprise RAG deployments, we recommend Azure AI Search if you're on Azure (managed, local region, hybrid search built-in) or Weaviate if you need multi-cloud flexibility or self-hosting for compliance.