[N] LangChain Interrupt 2026 announcements [N]

LangChain just wrapped Day 1 of Interrupt 2026 and announced a few things worth knowing about: **SmithDB** — A purpose-built distributed database for agent observability. The problem they're solving: agent traces are getting too large and complex for general-purpose databases. SmithDB is built with Rust, Apache DataFusion, and Vortex, designed specifically for multimodal content and long-span tracing. They're reporting P50 latency of 92ms for loading trace trees and 400ms for full-text search, with up to 12x speedup over previous LangSmith performance. Architecture is object storage + small Postgres metadata store + stateless services, so it scales elastically and can be self-hosted. **Context Hub** — A centralized system for managing agent context (AGENTS.md files, skills, policies, memory) in LangSmith. The interesting part is they're working with MongoDB, Pinecone, Elastic, and Redis on an open standard for agent memory — covering episodic, semantic, and procedural memory with versioning and portability across frameworks. **Deep Agents v0.6** — New release includes ContextHubBackend integration, an installable code interpreter that gives agents a programmable workspace inside the agent loop (distinct from sandboxes — this is for composing tools and managing state within the reasoning process), and you can scope specific file paths to different backends. The conference also has production case studies from Toyota, Coinbase, Lyft, LinkedIn, Bridgewater Associates, and others on deploying agents at enterprise scale. Andrew Ng keynoted alongside Harrison Chase.