MinimemoryInstall

Summary

Searches an in-memory vector database using vector similarity, BM25 keyword matching, or hybrid search, with optional metadata filtering.

Inputs

• databaseName (required) — The name of the in-memory database instance to query.
• searchMode — Search strategy: 'vector' (vector similarity), 'keyword' (BM25 full-text), or 'hybrid' (combined vector and keyword search).
• queryVector (required) — JSON array of numbers representing the query vector. Required when searchMode is 'vector' or 'hybrid'.
• keywords (required) — Search phrase or term for BM25 text search. Required when searchMode is 'keyword' or 'hybrid'.
• textFields — Comma-separated list of metadata key names to inspect during BM25 keyword search. Defaults to 'content,text,title,description'.
• topK — Maximum number of matching results to return. Defaults to 10.
• includeVectors — When enabled, includes the stored numerical vector array inside each returned item.
• minSimilarity — Filters out results with a similarity score below this threshold.
• useFilter — Enables metadata filtering on search results.
• metadataFilter — JSON object defining condition filters on vector metadata (e.g. {"category": "tech"} or {"score": {"$gt": 0.5}}).
• hybridAlpha — Balance weight between vector (1.0) and keyword (0.0) search results in hybrid mode. Defaults to 0.5.
• fusionMethod — Ranking fusion algorithm for hybrid search: 'rrf' (Reciprocal Rank Fusion) or 'weighted' score combination.
• bm25K1 — BM25 term frequency saturation parameter (0 to 3). Defaults to 1.2.
• bm25B — BM25 document length normalization parameter (0 to 1). Defaults to 0.75.

Output shape

A single object containing search execution details and a list of matching records in the results array.

The output object includes success, searchMode, k, filterApplied, resultsCount, and results. Each element inside results contains id, score, optional metadata, relevant rank/score metrics depending on search mode, and the vector array if includeVectors is true.

Examples

Example 1: Perform a vector similarity search using embeddings from a previous node

Set Database Name to 'docs', Search Mode to 'Vector Only', Query Vector to ={{ $json.embedding }}, and Number of Results (K) to 5.

Example 2: Execute a hybrid search combining vector embeddings and BM25 keywords with metadata filter

Set Search Mode to 'Hybrid', Query Vector to ={{ $json.embedding }}, Keywords to 'automation workflow', set Use Metadata Filter to true, and set Metadata Filter to {"status": "active"}.