Build Specialized Agents on a Stack You Control with NVIDIA NemoClaw

retained text

Building an autonomous agent requires a model, a harness, and domain-specific tools. To deploy an agent more securely, developers need a runtime that enforces policies, along with optimization and observability tools for reliable execution.
For production agents, the challenge lies in making these agents capable and safe to operate with access to real enterprise systems and data.
This post explains how these components work together, with a focus on how NVIDIA NemoClaw open blueprints enable developers to build specialized agents on a stack they can own, customize, and run. This post provides criteria to help developers integrate NemoClaw with a harness they already use: OpenClaw, Hermes Agent, or LangChain Deep Agents.
What an agent harness provides
An agent requires four things:
- A model that reasons.
- A harness that defines how the agent operates.
- Tools and skills that extend its capabilities.
- A runtime that governs what the agent can access and do in production.
NemoClaw works with agent harnesses
NemoClaw open blueprints currently support three harnesses:
- OpenClaw can access files and applications, write code, spawn sub-agents, and maintain context across sessions. It runs on your own machine or server and supports long-running sessions, memory, and scheduled tasks.
- Hermes Agent, from Nous Research, is designed to get better with use. As teams correct the agents’ approach and repeat workflows, Hermes captures those patterns as durable skills that can be reused in future work.
- LangChain Deep Agents provides a customizable harness for complex, multistep tasks. It combines planning, file-based context management, memory, and delegation to specialized subagents.
Each harness gives an agent skills, tools, orchestration, and memory.
Using NemoClaw for full-stack production agents
NemoClaw is not a competing harness, and it is not a runtime. It is a collection of blueprints that assembles an open agent stack on top of the harness selected by the developer. The stack combines model and harness optimization with runtime governance.
Each NemoClaw blueprint brings together four components:
- A tuned open harness, such as 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 can help improve throughput, cost, and reliability.
- NVIDIA OpenShell is the secure runtime that governs agent access to files, networks, credentials, and tools.
If you are already building with OpenClaw, Hermes Agent, or LangChain Deep Agents, a NemoClaw blueprint can assemble the surrounding stack without requiring you to rewrite the agent.
Why co-optimizing the model and harness matters
An agent combines a model and a harness. The model thinks, while the harness provides the environment in which it operates. When the two are matched, the model can focus on the task. When they aren’t, it may ask for context again, stop early, or loop without progress.
For example, LangChain tuned its Deep Agents harness for Nemotron 3 Ultra by adjusting prompts, tools, and middleware without retraining the model. Starting from an already strong baseline, the harness tuning brought Nemotron 3 Ultra to near-frontier performance at approximately 10x lower cost per run compared with leading closed models.
The NemoClaw blueprint for LangChain Deep Agents packages these optimizations directly. Every developer using LangChain Deep Agents with Nemotron 3 Ultra can access these optimizations without reproducing the tuning work. The tuned Nemotron 3 Ultra model profile is available through LangChain on GitHub.
When to use each
Use a harness such as OpenClaw, Hermes Agent, or LangChain Deep Agents to build and experiment with the agent’s skills, tools, memory, and orchestration logic.
Use a NemoClaw blueprint to assemble and run a specialized agent on an open stack that combines a tuned harness, models, optimization, telemetry, and OpenShell runtime controls.
A harness helps you build a capable agent. A NemoClaw blueprint helps you run that agent on an open stack with model and harness optimization, telemetry, and runtime governance.
Example: A memory-driven Chief of Staff, built on Hermes
The NemoClaw community repository includes a memory-driven Chief of Staff recipe. The agent maintains a persistent, human-readable model of the people, projects, priorities, and working patterns it encounters, so it doesn’t have to reconstruct context from scratch on every task. It is built as a deployable Hermes profile for NemoClaw, as outlined in the technical design walkthrough.
The recipe shows the stack working together rather than in isolation. OpenShell governs the agent’s file system, process, and network access, and keeps credentials for managed inference and Model Context Protocol (MCP) connections outside the sandbox. This helps keep the agent bound by policy even when its memory or retrieved context leads it toward an unintended action. NeMo Relay is integrated natively into the Hermes session. It closes trace spans at real session boundaries, including a new session, reset, or expiry, and exports the resulting trajectories for audit and further optimization.
Importance of agent governance
Every NemoClaw reference includes OpenShell, the runtime that adds security and policy enforcement around the agent harness and tools, governing the actions a model can take. The OpenShell tutorial covers how to create your first policy to prevent data exfiltration to GitHub. NemoClaw references include more sophisticated OpenShell policies that govern agent connections across many different internal tools, sites, and communication channels.
Deploy across cloud, data center, and local systems
NemoClaw agents can run in the cloud, in data centers, on NVIDIA DGX Spark, and on NVIDIA RTX PCs and workstations. A NemoClaw clinical agent for a hospital and a NemoClaw simulation agent for electronic design automation can run on the same stack while using entirely different tools and skills.
NemoClaw agents can also work alongside frontier agents. Claude, Codex, or Cursor agents can delegate domain-specific subtasks to NemoClaw-powered specialists running on open models and owned infrastructure.
Get started
Choose a NemoClaw blueprint for your harness, then customize its tools, models, and runtime policies for your workflow.
- Take the Deploy Secure Agents with NemoClaw DLI course—a 101-level introduction to the stack.
- Follow the NemoClaw quickstart to deploy an agent through a NemoClaw blueprint.
- Explore the NemoClaw GitHub repository for the full example and additional blueprints.
- Learn how to deploy self-evolving agents for faster, more secure research with Hermes Agent and NemoClaw.
Stay up to date on NVIDIA AI by subscribing to NVIDIA news and following NVIDIA AI on LinkedIn, X, Discord, and YouTube. Visit the developer page for resources to get started. Explore open Nemotron models and datasets on Hugging Face and NVIDIA Blueprints on build.nvidia.com. And engage with Nemotron livestreams, tutorials, and the developer community on NVIDIA forums and Discord.

document fetched 2026-10-31 02:54 · canonical https://developer.nvidia.com/blog/build-specialized-agents-on-a-stack-you-control-with-nvidia-nemoclaw