LangChain July 2026 Newsletter — NemoClaw Blueprint, OpenWiki Brains, and More

Jensen Huang and Harrison Chase discussed the need for open agent systems and unveiled the NVIDIA NemoClaw for LangChain Deep Agents blueprint. This blueprint aims to help teams build agent systems they can directly own, tune, and control.

On the LangSmith side, three updates were announced.

• LangSmith Sandboxes free trial — isolated environments where agents can safely run and test code
• LangSmith Fleet — connect agents in Slack with one click, supporting in-thread HITL approvals and document exchange
• Voice agent tracing — captures audio, STT/TTS latency, interruptions, and more in a single trace for voice agents built with Pipecat, LiveKit, OpenAI Realtime, and Gemini Live

On the open-source side, OpenWiki Brains stands out. It's a general-purpose memory layer that automatically converts and updates various sources—Gmail, Notion, git repositories, X, Hacker News, web search, and more—into a local wiki that agents can use.

Additionally, an integrated stack of Deep Agents and Harbor now enables agent evaluation through isolated sandbox execution, parallel trials, and LangSmith traces, and RLMs in Deep Agents can be used to break down large-context tasks into programmatic sub-agent calls.

LangChain Academy has launched a new introductory course, Introduction to Deep Agents. The Interrupt conference is set to expand to New York and London.