Run open-source Deep Agents Code, tuned for Nemotron 3 Ultra, to plan, edit and test code with enterprise governance.
LangChain Deep Agents Code (dcode) is an open source terminal coding agent built on the Deep Agents SDK. It is designed for complex software engineering tasks that require planning, file edits, command execution, test runs, and iterative review.
With NVIDIA NemoClaw, Deep Agents Code runs as a governed blueprint on NVIDIA Nemotron open models in an NVIDIA OpenShell runtime environment. Teams can point it at codebases while keeping source code, credentials, execution, and network access under enterprise control.
Deep Agents Code can break down complex engineering work, inspect a repository, make coordinated edits across files, run commands and tests, and return a diff for review. It is designed for multi-step engineering work like modernizing legacy codebases, upgrading dependencies and frameworks, repairing tests, and making coordinated code changes that need planning, execution, and review.
The blueprint pairs NVIDIA Nemotron 3 Ultra with LangChain's open-source Deep Agents Code harness. Teams get a model-neutral coding agent architecture with the flexibility to adapt the harness, model, and runtime as their needs change.
Deep Agents Code runs inside an OpenShell runtime environment with deny-by-default networking, approval controls, and an audit trail. Teams can allow the agent to work on sensitive codebases while keeping control over what resources it can access and execute.
Deep Agents manages long-running context through summarization, result offloading, persistent memory, skills, and subagents. Teams can encode repository conventions, migration playbooks, review standards, and domain-specific instructions so the agent improves on repeated work.