LangChain has introduced SmithDB, a distributed database designed to enhance agent observability, at the Interrupt! conference. Built with Rust, Apache DataFusion, and Vortex, SmithDB is tailored for managing large-scale, complex agent tracing data, including multimodal content and long-span scenarios. Its architecture features object storage, a small Postgres metadata store, and stateless services, allowing for elastic scaling and self-hosted deployment without local disk management. SmithDB demonstrates impressive performance metrics, with a P50 latency of 92ms for loading trace trees and 400ms for full-text search. It has improved core LangSmith experience speeds by up to 12 times. Currently, SmithDB is fully operational, handling 100% of U.S. cloud ingestion and tracing UI queries, with additional product integrations nearing completion.