Couchbase Query Vector Store icon

Couchbase Query Vector Store

Work with your data using the Couchbase Query Vector Store. This node conducts vector retrievals using the Query service with SQL++ vector indexes.

Couchbase Query Vector Store

AI-generated

Summary

Searches, inserts, retrieves, or updates documents in a Couchbase bucket, scope, and collection using vector embeddings and SQL++ vector indexes. Choose an operation mode, configure the required Couchbase location and embedding/text field keys, and connect an embedding model; retrieval can also use metadata filters or a reranker.

Inputs

  • Operation Mode (required) — Choose Get Many, Insert Documents, Retrieve Documents as a Vector Store, Retrieve Documents as an AI Agent Tool, or Update Documents.
  • Couchbase Bucket, Scope, and Collection (required) — Select each from the lists or enter its name. Scope and collection lists depend on the selected bucket and scope.
  • Distance Strategy — Similarity metric for vector search: Dot Product (default), Euclidean, Euclidean Squared, or Cosine.
  • Embedding Field Key (required) — Field containing each document's embedding array.
  • Text Field Key (required) — Field containing the document's text.
  • Embedding connection (required) — Connect an embedding model for document embedding and query search.
  • Get Many: Prompt, Limit, Include Metadata, Rerank Results, Metadata Filter — Prompt is required for Get Many. Limit defaults to 4 and Include Metadata defaults to true. Metadata filters are optional; enabling reranking requires a connected reranker.
  • Insert Documents: Document input, Embedding Batch Size, Add Vector Options — Connect a document source. Batch size defaults to 200 on node versions 1.1 and later. Optional vector options provide comma-separated IDs and a JSON array of metadata objects.
  • Retrieve as Vector Store: Rerank Results, Metadata Filter — Use as an AI vector store connection, optionally filter by metadata, and optionally connect a reranker. The node must be connected to a vector store retriever.
  • Retrieve as AI Agent Tool: Description, Limit, Include Metadata, Rerank Results, Metadata Filter — Tool description is required. Limit defaults to 4 and Include Metadata defaults to true. Metadata filtering and reranking are optional; reranking requires a connected reranker. For node versions up to 1.2, a tool name is also required.
  • Update Documents: ID — Required document ID to update. The incoming item is converted to a document; exactly one document per item is expected.

Output shape

Depends on the selected mode: a list of ranked document items, inserted or updated document items, an AI vector store connection, or an AI tool connection.

Get Many returns items containing document pageContent and score, with metadata included by default. Insert and Update return serialized document items; inserted IDs are added as documentId when returned by Couchbase. Retrieve modes expose a connection for downstream AI nodes. The AI Agent Tool expects a search query in its input field and returns matching document text, with metadata included by default.

Links

Discussion