Skip to content

Vector Store

Example Code: examples/knowledge/vectorstores

Vector store is the core component of the Knowledge system, responsible for storing and retrieving vector representations of documents.

Supported Vector Stores

trpc-agent-go supports multiple vector store implementations:

Vector Store Description
Memory In-memory vector store
PGVector PostgreSQL + pgvector extension
TcVector Tencent Cloud vector database
Elasticsearch Supports v7/v8/v9 versions
Qdrant High-performance vector database
Milvus High-performance vector database

Search Modes

Vector stores support four search modes. The system automatically selects the most appropriate mode based on the query:

Search Mode Enum Value Description Requirements
Vector SearchModeVector Semantic similarity search, understands query intent Requires Embedder
Keyword SearchModeKeyword Exact keyword matching, suitable for technical terms PGVector requires WithEnableTSVector(true)
Hybrid SearchModeHybrid Combines vector and keyword search (recommended, default) PGVector requires WithEnableTSVector(true)
Filter SearchModeFilter Metadata-only filtering, no similarity score -

Examples: 

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
// Default automatic search
reqSearch := &knowledge.SearchRequest{
    Query: "Large language model applications",
    //History: nil,
    //UserID: "",
    //SessionID: "",
    //MaxResults: 0,
    //MinScore: 0,
    //SearchFilter: nil,
    //SearchMode: 0,
}
rspSearch, err := kb.Search(ctx, reqSearch)
if err != nil {
    log.Debugf("Failed to search knowledge base: %v", err)
    return
}
log.Infof("Search Result:[%+v]", rspSearch)
for i, r := range rspSearch.Documents {
    log.Infof("rspSearch.DOC_%d.score=%v,doc:[%+v]", i, r.Score, r.Document)
}

// Search with filter conditions
filter := knowledge.SearchFilter{
    //DocumentIDs: nil,
    //Metadata: nil,
    FilterCondition: searchfilter.And(
        searchfilter.Equal("category", "film"),
        searchfilter.Equal("country_code", "us"),
    ),
}
reqSearchWithFilter := &knowledge.SearchRequest{
    Query:        "Top ranked movie",
    SearchFilter: &filter,
}
rspSearchWithFilter, err := kb.Search(ctx, reqSearchWithFilter)
if err != nil {
    log.Debugf("Failed to search knowledge base: %v", err)
    return
}
log.Infof("Search Result:[%+v]", rspSearchWithFilter)
for i, r := range rspSearchWithFilter.Documents {
    log.Infof("rspSearchWithFilter.DOC_%d.score=%v,doc:[%+v]", i, r.Score, r.Document)
}

Filter Support

All vector stores support filter functionality, including ID filtering, metadata filtering, and complex condition filtering (FilterCondition).

More Content

  • Memory - In-memory vector store configuration
  • PGVector - PostgreSQL + pgvector configuration
  • TcVector - Tencent Cloud vector database configuration
  • Elasticsearch - Elasticsearch configuration
  • Qdrant - Qdrant configuration
  • Milvus - Milvus configuration