Vector Search

Qdrant Integration

qdrant.go provides the vector database client via gRPC:

Collections

// Create a new collection
rag.CreateCollection(ctx, client, "docs", vectorSize)

// Check if collection exists
exists := rag.CollectionExists(ctx, client, "docs")

Upsert

// Store embedded chunks
rag.Upsert(ctx, client, "docs", points)

Search

results, _ := rag.Search(ctx, client, "docs", queryVector, limit)

Ollama Embeddings

ollama.go generates vector embeddings using Ollama's embedding API:

embedding, _ := rag.Embed(ctx, ollamaClient, "nomic-embed-text", text)

Query Interface

query.go provides the high-level search API:

cfg := &rag.QueryConfig{
    Collection: "docs",
    Model:      "nomic-embed-text",
    Limit:      5,
    Format:     "text", // or "json", "xml"
}

results, _ := rag.Query(ctx, cfg, "How does authentication work?")

Output Formats

text: Plain text with source attribution
json: Structured JSON for programmatic use
xml: XML format for MCP tool responses

Ingestion Pipeline

ingest.go provides end-to-end document processing:

cfg := &rag.IngestConfig{
    Collection: "docs",
    Model:      "nomic-embed-text",
    ChunkSize:  512,
}

// Single file
rag.IngestFile(ctx, cfg, "/path/to/README.md")

// From reader
rag.Ingest(ctx, cfg, reader, "source-name")

The pipeline: read file → chunk → embed each chunk → upsert to Qdrant.