Dev Tools

Plot twist: Weaviate v1.30 quietly made hybrid search the default behavior instead of an opt-in flag. Combined with the built-in embedding modules, you can stand up a real RAG pipeline without writing a line of embedding code.

The Setup

Spin up Weaviate, tell it which embedding module to use (`text2vec-openai`, `text2vec-cohere`, `text2vec-transformers`), and it handles the vectorisation on ingest and query. No more two-step embed-then-upsert dance.

{`# docker-compose snippet
services:
  weaviate:
    image: cr.weaviate.io/semitechnologies/weaviate:1.30.0
    environment:
      DEFAULT_VECTORIZER_MODULE: text2vec-transformers
      ENABLE_MODULES: text2vec-transformers,reranker-transformers
      TRANSFORMERS_INFERENCE_API: http://t2v-transformers:8080
    ports:
      - "8080:8080"`}

The Money Pattern

The new default hybrid query blends vector similarity with BM25, weighted by an `alpha` parameter. 0.5 is a sane default — pure semantic on one end, pure keyword on the other. GraphQL is still Weaviate's love language and it shows.

{`{
  Get {
    Claim(
      hybrid: {
        query: "roof leak after hailstorm in Brisbane"
        alpha: 0.5
      }
      where: { path: ["region"], operator: Equal, valueText: "qld" }
      limit: 5
    ) {
      description
      region
      _additional { score explainScore }
    }
  }
}`}

The Catch

Weaviate is heavier than Chroma or LanceDB. You're running a JVM-adjacent service, you're learning their schema DSL, and the GraphQL API is great until you need to debug it at 1am. Ops cost is real — this isn't a "drop it in a Lambda" tool.

The Verdict

For teams already running real infra who want hybrid search and embedding pipelines as one product, v1.30 makes Weaviate the most batteries-included vector DB on the market. If you're solo on a laptop, it's overkill. If you're shipping a serious search product, it's worth the JVM tax.