LangChain Ships LangSmith AI Search, Managed Deep Agents v0.9, and Engine v2 at Interrupt

LangChain announced a suite of new features at Interrupt NYC & London aimed at helping teams "own their intelligence" by controlling the model, harness, and context for their agents. The updates focus on improving observability, agent management, and open-source integration.

New LangSmith Features

The platform introduced several tools to enhance agent debugging and security:

• LangSmith AI Search: Enables semantic search of traces using natural language questions (e.g., "did PII leak?") instead of traditional filters.
• Managed Deep Agents v0.9: Allows building dynamic agents with user-level memory, Slack integration, task scheduling, and free built-in web search.
• LangSmith Engine v2: Adds Red Teaming to catch issues pre-production, detects inefficient agent work, and automatically validates proposed fixes.
• LangSmith Trajectories: Provides a chronological view of agent paths across main agents and subagents.
• LangSmith Custom Apps: Lets users build and publish custom interfaces directly within LangSmith.
• LangSmith Fine-Tuning: Converts production trajectories into training data for fine-tuning models via Baseten or Fireworks.

Open Source Updates

LangChain released v1.4.0, which adds support for FastMCP through the langchain.mcp namespace. Additionally, the team highlighted how LangGraph orchestrates Jev (TypeSafe AI’s decision model) to build faster and cheaper production agents.

Agentic Operating Model

LangChain is promoting its Agentic Operating Model framework, which guides teams on scaling production AI through continuous evaluation and governance. Virtual sessions are scheduled for Oct 29 (EMEA) and Nov 5 (North America).