NemoClaw + dcode: A governed blueprint for AI coding agents
LangChain8 July 2026Watch on YouTube
Description
In this demo, we run dcode (LangChain's Deep Agents Code) inside a NemoClaw-managed OpenShell sandbox, with Nemotron 3 Ultra served through Baseten for inference. We walk through onboarding, provisioning the sandbox, handing dcode a failing test to fix, and then inspecting the sandbox's policy and logs to see how teams get auditability without changing the developer's normal workflow. Chapters: 0:00 dcode running inside a governed sandbox 0:39 Installing NemoClaw 0:51 Onboarding and picking the dcode integration 0:59 Configuring the inference provider 1:39 Naming and reviewing the sandbox 1:56 Setting resource profile and policy tier 2:26 Provisioning the OpenShell sandbox 2:40 Connecting to the sandbox 3:00 Setting up a failing test 3:44 Handing the failing test to dcode 4:41 Verifying the fix 5:28 Inspecting policy and logs 6:40 Wrap up Resources: → LangChain Deep Agents: https://www.langchain.com/deep-agents → LangSmith: https://www.langchain.com/langsmith → LangChain Academy: https://academy.langchain.com
What you'll learn
- NemoClaw provides a governed sandbox environment where AI coding agents like dcode can work safely with full governance and auditability.
- dcode automatically fixes failing tests by modifying code, with all actions traceable in the sandbox logs.
- Resource profiles and policy tiers control what actions an AI coding agent can perform, without disrupting developers' normal workflow.
- Nemotron 3 Ultra can be configured as an inference provider via Baseten to power dcode.
- Teams can inspect sandbox policies and logs to gain full transparency into what AI agents do in code development.