LangChain announces at Interrupt 2026

LangChain unveiled new features spanning agent observability, memory, and execution environments on Interrupt 2026 Day 1.

• SmithDB is a distributed DB for agent traces. Built with Rust, Apache DataFusion, and Vortex, it targets multimodal content and long-term tracking. It presents tree loading at P50 92ms, full-text search at 400ms, and up to 12x speed compared to the existing LangSmith. Its architecture consists of object storage + small Postgres metadata + stateless services, and self-hosting is also possible.

• Context Hub is a system for centrally managing AGENTS.md, skills, policies, and memory within LangSmith. It is pushing an open standard for agent memory together with MongoDB, Pinecone, Elastic, and Redis, with episodic / semantic / procedural memory, version control, and cross-framework portability in scope.

• Deep Agents v0.6 adds ContextHubBackend integration, an installable code interpreter, and backend scope configuration by file path. The code interpreter is designed as a workspace that composes tools and handles state within the reasoning loop, separate from the sandbox.

At the event, agent operation case studies from Toyota, Coinbase, Lyft, LinkedIn, Bridgewater Associates, and others were also introduced, and Andrew Ng and Harrison Chase delivered the keynotes.