LangChain Interrupt 2026: LangSmith Engine, SmithDB, Managed Deep Agents, and LangChain Labs Ship

LangSmith Engine entered public beta — automates production debugging from trace to fix PR

Monitors live production traces, clusters failures into named issues, diagnoses root causes, and drafts code fix pull requests. Early adopters Cogent and Campfire used it to resolve issues affecting thousands of traces.

SmithDB is a Rust-based, agent-native database delivering up to 15x faster LangSmith performance

Built on Apache DataFusion and the Vortex columnar format, backed by object storage with a Postgres metastore. P50 trace tree load is 92ms; P50 single run load is 71ms. Engineered specifically for agent-native query patterns rather than adapted from general-purpose databases.

SmithDB P50 benchmarks: 92ms trace tree load, 71ms single run load — up to 15x faster than prior LangSmith storage

These figures represent the performance envelope for the new agent-native database under typical workloads, establishing a concrete baseline against which future iterations can be measured.

Managed Deep Agents launched as a hosted API runtime for durable, long-horizon agent workloads

Accessible via the /v1/deepagents API endpoint. Completes LangChain's fully hosted stack for teams that want to run autonomous agents without managing their own execution infrastructure.

LLM Gateway added spend limits and PII redaction before requests leave the customer environment

Addresses a core enterprise governance requirement: sensitive data is stripped locally before any LLM call is made, reducing compliance exposure. Combined with Context Hub for versioning agent instructions, these features signal a deliberate governance and auditability push.

Context Hub enables versioning and management of agent instructions across deployments

Gives teams a structured way to track changes to agent prompts and configurations over time — important for debugging regressions and maintaining auditability in production environments.

Sandboxed code execution reached general availability at Interrupt 2026

Moves from beta to GA, making isolated code execution a production-ready capability within the LangChain platform.

Messages View released as a new interface for inspecting agent conversation threads

Provides a structured view of message-level interactions within agent runs, aiding debugging and evaluation workflows inside LangSmith.

LangChain Labs launched at Interrupt as a continual-learning research initiative for open, applied agent research

Launch partners are Harvey, NVIDIA, Prime Intellect, Fireworks, and Baseten. The announcement does not elaborate on each partner's specific role in the initiative.

LangChain Labs launch partners include Harvey, NVIDIA, Prime Intellect, Fireworks, and Baseten

Five organizations are named as partners on the LangChain Labs initiative. The source lists them without describing their individual contributions or domains.

All 23 Interrupt 2026 sessions are now available on demand

Featured teams include Cisco Customer Experience, LinkedIn, and Rippling, representing enterprise adoption of LangChain tooling at scale. On-demand availability extends the conference's reach beyond attendees.

LangChain's Interrupt announcements collectively represent a shift from framework provider to full managed platform with a research arm

The four pillars — automated production debugging, agent-native storage, hosted execution, and applied research — cover the build, run, observe, and improve cycle. The addition of LangChain Labs with blue-chip AI partners adds a longer-horizon dimension to what had been a primarily tooling-focused company.