Skip to content
MilvusNode SDK
Sections

Hybrid search

Markdown

Copy for agent

Connection and CORS help

Use HTTPS for remote clusters or HTTP on loopback. Requests go directly from your browser to the endpoint. Run List collections to populate the collection selector. When copying an authenticated cURL command, set MILVUS_TOKEN in your shell.

For cross-origin requests, configure the endpoint to answer OPTIONS and allow the docs origin, POST, Content-Type, and Authorization if you use a token. Try the copied cURL command to troubleshoot outside the browser.

POST/v2/vectordb/entities/hybrid_search

{
  "dbName": "default",
  "collectionName": "books",
  "search": [
    {
      "annsField": "vector",
      "data": [
        [
          0.1,
          0.2,
          0.3,
          0.4
        ]
      ],
      "limit": 5
    },
    {
      "annsField": "vector",
      "data": [
        [
          1,
          0,
          0,
          0
        ]
      ],
      "limit": 5
    }
  ],
  "rerank": {
    "strategy": "rrf",
    "params": {
      "k": 60
    }
  },
  "limit": 3,
  "outputFields": [
    "title"
  ]
}

Example response

{
  "code": 0,
  "data": [
    {
      "id": 101,
      "title": "Invisible Cities",
      "distance": 0.033
    }
  ],
  "cost": 0,
  "topks": [
    1
  ]
}
cURL command
curl --request POST --url 'https://your-cluster.example.com/v2/vectordb/entities/hybrid_search' \
  --header 'Content-Type: application/json' \
  --header "Authorization: Bearer ${MILVUS_TOKEN}" \
  --data-raw '{
  "dbName": "default",
  "collectionName": "books",
  "search": [
    {
      "annsField": "vector",
      "data": [
        [
          0.1,
          0.2,
          0.3,
          0.4
        ]
      ],
      "limit": 5
    },
    {
      "annsField": "vector",
      "data": [
        [
          1,
          0,
          0,
          0
        ]
      ],
      "limit": 5
    }
  ],
  "rerank": {
    "strategy": "rrf",
    "params": {
      "k": 60
    }
  },
  "limit": 3,
  "outputFields": [
    "title"
  ]
}'

Combine multiple vector searches and rerank the candidates into one result list.

Request fields

dbName string
Database name. Operations on collections use default when omitted.
collectionName string (required)
Name of the collection.
partitionNames array of string
Names of the partitions to target.
search array of object
Sub-searches with query data, vector field and limit.
rerank object
Reranking strategy and parameters, such as reciprocal rank fusion (rrf).
limit integer
Maximum number of rows or hits per query input.
offset integer
Number of matching results to skip.
groupingField string
Scalar field used to group search hits.
groupSize integer
Number of hits requested per group.
strictGroupSize boolean
Require the specified number of hits in each result group.
outputFields array of string
Field names to include in the returned entities.
consistencyLevel string
Read consistency level, such as Strong, Session, or Bounded.
functionScore object
Scoring functions and parameters to apply to candidates.
functionChains array of object
Ordered function chains to apply during search.

See HttpClient.hybridSearch for the Node.js SDK method.