LangChain Adds AI Search, Fine-Tuning and Custom Apps to LangSmith
LangChain used its Interrupt events in NYC and London to launch a batch of LangSmith features, including semantic trace search, a second version of Engine, and fine-tuning tools. According to the company, the goal is to help teams "own their intelligence" by giving them more control over how they build, understand and improve agents in production.
- LangChain launched LangSmith AI Search, Engine v2, Trajectories, Custom Apps and Fine-Tuning at Interrupt NYC and London.
- LangSmith Fine-Tuning turns production trajectories into training data and works with Baseten or Fireworks.
- Managed Deep Agents v0.9 includes built-in web search for free.
- LangChain v1.4.0 supports FastMCP through the langchain.mcp namespace.
- Engine v2 adds Red Teaming and automatically tests and validates its proposed fixes.
New LangSmith features for production agents
LangChain's October 2026 newsletter lists six additions to LangSmith, its agent engineering platform.
- LangSmith AI Search: lets users find conversations through a semantic understanding of their traces. The company gives example questions such as "did the user get frustrated?" or "did PII leak?", asked without relying on traditional filters.
- Managed Deep Agents v0.9: for building dynamic, personalized agents tailored to each run's context. According to LangChain, these agents can react in Slack, have user-level memory and tool use, and schedule tasks. Built-in web search is included for free.
- LangSmith Engine v2: adds Red Teaming to catch issues before they reach production. LangChain says it also detects harder-to-spot problems such as inefficient agent work and metric trends, and automatically tests and validates its proposed fixes.
- LangSmith Trajectories: offers a chronological view of each session across the main agent and its subagents, so developers can follow an agent's path through trace data.
- LangSmith Custom Apps: allows teams to build, publish and run custom interfaces directly in LangSmith, turning recurring workflows into shared tools built on their LangSmith data.
- LangSmith Fine-Tuning: turns production trajectories into training data, fine-tunes models with Baseten or Fireworks, then evaluates and serves the tuned models.
Open source updates and learning resources
On the open source side, LangChain v1.4.0 now supports FastMCP through the langchain.mcp namespace. The newsletter also points to a write-up on building production agents with Jev and LangGraph. It shows how LangGraph orchestrates Jev, TypeSafe AI's decision model, to build agents that are faster and cheaper to run, according to LangChain.
The company also promotes The Agentic Operating Model, a framework on how leading teams bring people, processes and technology together to scale production AI. It covers rethinking the agent development lifecycle, building continuous evaluation practices and embedding governance. Virtual sessions were scheduled for EMEA on Oct 29 at 10:00 GMT / 11:00 CET and for North America on Nov 5 at 10:00 AM PT / 1:00 PM ET.
Two live workshops under the Speak the Lang banner were also announced. Agent Evals 101 on Oct 28 covers measuring agent performance by assessing its execution trajectory, meaning the sequence of messages and tool calls it produces. Agent Memory 101 on Nov 3 looks at context engineering and how Deep Agents, LangChain's open source agent harness, helps keep context windows lean.
Events and customer example
LangChain listed several upcoming events it is hosting, including a hackathon with Data Reply in Munich on Oct 27 on building with agent harnesses, and a LangSmith Roadshow in Chicago on the same day. Other dates include a meetup in Vienna, VA on Oct 29 and a session in Toronto on Nov 4 about owning intelligence with open-weight models. A community event on agents in production was listed for Mexico City on Oct 29.
As a customer example, the newsletter cites Zip. Before, adding a feature to an early version of its chatbot or agent took weeks, and its in-house traces pipeline was missing details the team needed. According to LangChain, LangSmith and LangGraph now make shipping new features faster.
What it means for teams building agents
The releases cover several stages of the agent lifecycle: finding problems in traces, testing for issues before deployment, and feeding production data back into model training. For teams already running agents on LangSmith, the fine-tuning option with Baseten or Fireworks and the semantic search could shorten the loop between spotting a problem and fixing it. The newsletter does not give pricing or availability details for the new features, apart from noting that web search in Managed Deep Agents is free.
