OpenSearch k-NN 向量检索索引与查询实战

索引与映射

PUT /vectors
{
  "settings": { "index.knn": true },
  "mappings": {
    "properties": {
      "title": { "type": "text" },
      "vec": { "type": "knn_vector", "dimension": 768 }
    }
  }
}

插入

POST /vectors/_doc
{ "title": "doc1", "vec": [0.12, 0.03, ...] }

查询（k-NN）

POST /vectors/_search
{
  "knn": {
    "field": "vec",
    "query_vector": [0.11, 0.02, ...],
    "k": 10,
    "num_candidates": 100
  }
}

总结

启用 index.knn 并使用 knn_vector 映射即可进行向量检索，num_candidates 影响召回与性能。