Integrations

Weaviate (vector search)

Connect Weaviate as a vector provider for the Semantic Search built-in tool. Weaviate is not a standalone integration — it is one provider option within the Semantic Search tool, which is available to both Agents and Flows. The Semantic Search tool also supports the platform’s built-in pgvector and Cloudflare Vectorize as vector providers.

What is Weaviate?

Weaviate is a vector database for semantic search. Use it for RAG (Retrieval Augmented Generation) at scale.

When to use Weaviate vs the platform store

  • Use the platform vector store (pgvector): Fastest to get started, no external account, good for getting going and for moderate volumes.
  • Use Weaviate (or Cloudflare Vectorize): When you want a dedicated vector database, advanced filtering, or to run at larger scale.

Setting up Weaviate

  1. Create a Weaviate cluster at console.weaviate.cloud
  2. Get your cluster URL and API key
  3. Provide those credentials when you configure the Weaviate provider on a Semantic Search tool (see below)

Using Weaviate with the Semantic Search tool

Semantic Search generates query embeddings with OpenAI, so you need an OpenAI API key configured before you use it. Weaviate is then configured as the vector provider when adding the Semantic Search tool to an Agent or Flow:

  1. Add the Semantic Search tool
  2. Under Vector Provider, select Weaviate (the other options are pgvector and Cloudflare Vectorize)
  3. Enter your Weaviate cluster URL and API key, and the Collection Name (the Weaviate class to search)
  4. Optionally configure the similarity threshold and display name

The tool supports multiple instances, so you can add several Semantic Search tools pointing to different Weaviate collections in the same Agent.

RAG with Weaviate

Build retrieval-augmented generation by enabling the Semantic Search tool (with the Weaviate provider) on an Agent, or by adding a Vector Search step to a Flow. Either way, the relevant documents from the Weaviate collection become context for generating answers.

Next steps