AgentDB Advanced Features

Distributed vector search in AgentDB uses QUIC synchronization to coordinate multiple nodes with sub-millisecond latency. Enable QUIC by configuring `enableQUICSync: true`, specifying a sync port, and listing peer addresses. This allows agents to perform hybrid searches across the distributed cluster while maintaining real-time consistency.

Hybrid search combines vector similarity matching with metadata filtering for precise, contextual results. Standard vector similarity returns matches by embedding distance alone. AgentDB's hybrid approach lets you filter by metadata fields—like year or category—alongside vector scores, reducing irrelevant results in large datasets.

Yes, AgentDB supports custom distance metrics including Cosine Similarity, Euclidean Distance, and Dot Product, plus the ability to implement your own. Custom metrics allow you to tailor vector similarity scoring to your domain's specific requirements and optimize relevance for specialized use cases.

MMR retrieves diverse results by balancing relevance with novelty, avoiding redundant or similar items in the result set. Instead of returning the top matches by similarity alone, MMR penalizes candidates that are too similar to already-selected results, providing richer exploration of your vector space.

Multi-database sharding in AgentDB enables horizontal scalability by splitting data across databases by domain or shard key. For massive datasets or high-throughput clusters, sharding distributes query load and storage, supporting production-scale deployments without single-node bottlenecks.

QUIC synchronization in AgentDB achieves sub-millisecond latency (<1ms) between nodes with built-in encryption and multiplexing. This enables real-time coordination for distributed AI systems, making it suitable for time-sensitive workloads requiring immediate cross-node consistency.