NVIDIA NemoClaw blueprints bundle Nemotron, NeMo tools and OpenShell for three agent harnesses

NVIDIA has released NemoClaw, a collection of open blueprints that assemble a full agent stack around a harness the developer already uses. The blueprints currently support three harnesses: OpenClaw, Hermes Agent and LangChain Deep Agents. NVIDIA announced it in a developer blog post dated Oct 30, 2026, written by Nicola Sessions.

What NemoClaw is

NVIDIA states that NemoClaw is not a competing harness and not a runtime. It is a set of blueprints that builds an open agent stack on top of the harness the developer selects. Each blueprint combines four components:

• A tuned open harness (Hermes Agent, LangChain Deep Agents or OpenClaw), optimized to work with NVIDIA Nemotron.
• Nemotron open models.
• NVIDIA NeMo tools for optimization, including NVIDIA NeMo Relay for observing and improving agent execution, plus routing capabilities that NVIDIA says can help improve throughput, cost and reliability.
• NVIDIA OpenShell, the secure runtime that governs agent access to files, networks, credentials and tools.

According to NVIDIA, developers already building on one of the supported harnesses can use a blueprint to assemble the surrounding stack without rewriting the agent.

Supported harnesses

• OpenClaw can access files and applications, write code, spawn sub-agents and maintain context across sessions. It runs on the developer's own machine or server and supports long-running sessions, memory and scheduled tasks.
• Hermes Agent, from Nous Research, captures corrected approaches and repeated workflows as durable skills that can be reused in future work.
• LangChain Deep Agents combines planning, file-based context management, memory and delegation to specialized subagents.

Nemotron 3 Ultra harness tuning

The post cites one concrete result. LangChain tuned its Deep Agents harness for Nemotron 3 Ultra by adjusting prompts, tools and middleware, without retraining the model. According to NVIDIA, starting from an already strong baseline, the tuning brought Nemotron 3 Ultra to near-frontier performance at approximately 10x lower cost per run compared with leading closed models. No benchmark names or scores are given in the post.

The NemoClaw blueprint for LangChain Deep Agents packages these optimizations, and the tuned Nemotron 3 Ultra model profile is available through LangChain on GitHub.

Runtime governance and deployment

NVIDIA says OpenShell keeps agents bound by policy even when memory or retrieved context suggests unintended actions. NemoClaw agents run across cloud, data center, NVIDIA DGX Spark, and NVIDIA RTX PCs and workstations. They can operate alongside frontier agents such as Claude, Codex or Cursor.

Getting started

NVIDIA points developers to three resources: the Deploy Secure Agents with NemoClaw DLI course (a 101-level introduction), the NemoClaw quickstart for deploying an agent through a blueprint, and the NemoClaw GitHub repository for examples and additional blueprints.

What this means

NemoClaw packages model choice, harness tuning, telemetry and policy enforcement as one reference stack rather than separate integration work. The LangChain result, as NVIDIA reports it, shows the approach in practice: the gain came from adjusting prompts, tools and middleware, not from retraining the model. The reported cost-per-run comparison comes from NVIDIA's post and is stated without published benchmark detail, so teams should measure it against their own workloads. Support for existing harnesses means the entry point is a blueprint layered over current agent code, with OpenShell supplying the policy layer over files, networks, credentials and tools.