Vector Stores

Fast similarity search with pluggable vector store backends.

Vector stores provide fast similarity search over document embeddings. AgentOven ships with two OSS vector store drivers and a pluggable registry for enterprise backends.

OSS Vector Store Drivers

Embedded (In-Memory)

Zero-dependency, brute-force cosine similarity search. Perfect for local development and small datasets. Supports up to 50,000 documents. Data is not persisted across restarts.

• No external dependencies — works out of the box
• Cosine similarity ranking
• 50,000 document limit per kitchen
• Metadata filtering support
• Ideal for development and testing

pgvector (PostgreSQL)

Full-featured vector search using the pgvector extension for PostgreSQL. Provides HNSW and IVFFlat indexing for production-scale similarity search. Tables are auto-created on startup.

• Production-ready with PostgreSQL
• HNSW and IVFFlat index support
• Persistent storage across restarts
• Scales to millions of vectors
• Bring your own PG + pgvector extension

Configuration

The embedded vector store is always available. To enable pgvector, set the connection URL:

# Embedded store is always available (default)

# Enable pgvector (requires PostgreSQL + pgvector extension)
$ export AGENTOVEN_PGVECTOR_URL="postgres://user:pass@localhost:5432/agentoven"

# AgentOven auto-creates the vector tables on startup

VectorStoreDriver Interface

All vector stores implement a common interface, making it easy to swap backends:

type VectorStoreDriver interface {
    Kind() string
    Upsert(ctx context.Context, kitchen string, docs []models.VectorDoc) error
    Search(ctx context.Context, kitchen string, vector []float64, topK int, filter map[string]string) ([]models.SearchResult, error)
    Delete(ctx context.Context, kitchen string, ids []string) error
    Count(ctx context.Context, kitchen string) (int, error)
    HealthCheck(ctx context.Context) error
}

API Endpoints

Enterprise Extensions

AgentOven Pro adds enterprise-grade vector store backends:

• Pinecone — managed vector database
• Qdrant — high-performance open-source
• Azure Cosmos DB — globally distributed vector search
• Chroma — embedded AI-native vector database
• Snowflake Cortex — data warehouse vector search