AI & LLM

Vector Database Search Query JSON Example

A JSON example of a vector database query and response — includes the query vector, topK, metadata filter, and ranked matches with scores. Copy-ready for Pinecone, Qdrant, and pgvector.

JSON

Generate mock data Open in Formatter

{
  "matches": [
    {
      "id": "doc_4821#chunk_3",
      "score": 0.8731,
      "metadata": {
        "title": "Resetting your password",
        "source": "help-center/account.md"
      }
    },
    {
      "id": "doc_1190#chunk_0",
      "score": 0.7402,
      "metadata": {
        "title": "Password requirements",
        "source": "help-center/security.md"
      }
    }
  ],
  "namespace": "help-center"
}

Field Reference

matchesrequiredarray<object>Nearest-neighbour results, ordered by score descending

matches[].idrequiredstringIdentifier of the stored vector/record

matches[].scorerequirednumberSimilarity score (cosine/dot-product); higher means more similar

matches[].metadataoptionalobjectAttributes stored with the vector — used for display, citation, and filtering

namespaceoptionalstringLogical partition the query ran against

Variants

Query requestSearch by an embedding vector, limit results with topK, and filter on metadata.

Query request

Generate mock data Open in Formatter

{
  "vector": [
    0.0023064255,
    -0.009327292,
    0.015797347,
    -0.0073122089
  ],
  "topK": 5,
  "includeMetadata": true,
  "filter": {
    "source": {
      "$eq": "help-center/account.md"
    }
  },
  "namespace": "help-center"
}

Upsert requestInsert or update vectors with their metadata.

Upsert request

Generate mock data Open in Formatter

{
  "vectors": [
    {
      "id": "doc_4821#chunk_3",
      "values": [
        0.0023,
        -0.0093,
        0.0157
      ],
      "metadata": {
        "title": "Resetting your password"
      }
    }
  ],
  "namespace": "help-center"
}

Common Use Cases

→Querying a vector database for semantically similar documents
→Powering retrieval in a RAG pipeline with metadata filtering
→Building recommendation and deduplication features on embeddings

vector databasepineconeqdrantsemantic searchsimilarityrag

Validate or format this JSON

One click loads this exact example into the tool — no copy-paste needed. Format it, validate it, explore the tree, or generate TypeScript types instantly.

Format Validate Explore tree Generate TypeScript Minify

Frequently Asked Questions

topK is the number of nearest neighbours to return — e.g. topK: 5 returns the five most similar vectors by score. For RAG you typically retrieve a small topK (3–10), then optionally re-rank before building the LLM context.

You attach metadata (source, date, tenant, access level) to each vector at upsert time, then pass a filter in the query so only matching vectors are considered. This enables multi-tenant isolation, recency filters, and permission-aware retrieval.

It depends on the metric. Cosine similarity ranges from -1 to 1 (usually 0 to 1 for text embeddings); dot product is unbounded; Euclidean distance is 0+ where lower is closer. Always check which metric your index uses before thresholding.

Related JSON Examples

AI & LLMEmbedding Vector ResponseEmbeddings API response with the vector array, model, and usage.AI & LLMRAG Document ChunkA RAG chunk with text, source metadata, embedding reference, and relevance score.API DesignSearch API ResponseSearch API response JSON with scored hits, highlights, facets, and pagination.