LangChain Interrupt 2026: SmithDB, LangSmith Engine, 9 Major Releases
LangChain's Interrupt 2026 conference produced nine infrastructure releases. SmithDB is a purpose-built distributed database for agent observability, written in Rust on Apache DataFusion and Vortex file format, delivering sub-second write latency and 12× performance improvement versus previous LangSmith infrastructure. LangSmith Engine provides an always-on agent self-improvement loop: every trace is analyzed for errors and product insights, with auto-generated PRs and evals gated by human approval.
Why It Matters
SmithDB addresses the core LLMOps gap — agent traces contain unbounded payloads that traditional databases fail to ingest quickly. LangSmith Engine marks the beginning of continual learning at the harness level, where agent improvement becomes automated and persistent rather than manual.