๐ฉNVIDIA Developer BlogโขStalecollected in 31m
Deploy Secure Self-Evolving Agents with NVIDIA NemoClaw
#ai-agents#data-security#enterprise-ai#open-sourcehermes-agent-and-nvidia-nemoclawnvidiahermes agentnemoclawopenshell
๐กLearn how to build secure, self-evolving AI agents that safely bridge internal data with public research sources.
โก 30-Second TL;DR
What Changed
Integrates Hermes Agent with NVIDIA NemoClaw for automated research workflows.
Why It Matters
This framework provides a blueprint for enterprises to build secure, agentic workflows that bridge the gap between private internal knowledge and public research data. It significantly reduces the security risks typically associated with deploying autonomous agents in corporate environments.
What To Do Next
Clone the open-source repository and test the NemoClaw integration with your internal GitHub or Slack data to evaluate its security policy enforcement.
Who should care:Developers & AI Engineers
Key Points
- โขIntegrates Hermes Agent with NVIDIA NemoClaw for automated research workflows.
- โขSupports data synthesis across enterprise tools including Outlook, Slack, and GitHub.
- โขUtilizes NVIDIA OpenShell to maintain security compliance for sensitive internal data.
๐ง Deep Insight
Web-grounded analysis with 15 cited sources.
๐ Enhanced Key Takeaways
- โขNVIDIA NemoClaw is an open-source reference stack built on OpenClaw, designed for running always-on AI agents securely within NVIDIA OpenShell containers, providing lifecycle management, onboarding, and agent operations with policy-based privacy and security guardrails.
- โขNVIDIA OpenShell is an open-source security runtime specifically for AI agent deployments, enforcing security policies at the execution layer using YAML configurations and kernel-level isolation technologies like Landlock, seccomp, and network namespaces for sandboxing.
- โขHermes Agent, developed by Nous Research, is a self-improving AI agent that can run locally on NVIDIA RTX PCs and DGX Spark, featuring self-evolving skills, persistent memory across sessions, and integration with various messaging platforms.
- โขNemoClaw supports diverse inference providers, including NVIDIA Endpoints, OpenAI, Anthropic, Google Gemini, and local models like Ollama, routing agent traffic through an OpenShell gateway while keeping credentials on the host.
- โขLeading enterprise software companies such as Siemens, Synopsys, Foxconn, Cadence, and Dassault Systรจmes are integrating NVIDIA NemoClaw and OpenShell to develop autonomous AI engineers and agents for complex workflows in sectors like semiconductor design, healthcare, and manufacturing.
๐ ๏ธ Technical Deep Dive
- NVIDIA NemoClaw serves as an open-source reference stack for deploying AI agents within NVIDIA OpenShell sandboxes.
- OpenShell creates individual, isolated sandboxes for each agent and sub-agent, specifically designed for AI that can modify its own environment.
- Security enforcement occurs at the kernel level, utilizing Landlock LSM for filesystem isolation, Seccomp BPF for syscall filtering, and Open Policy Agent (OPA) for network policies.
- The filesystem is strictly controlled, allowing agents to write only to designated directories (e.g.,
~/.hermes/,/sandbox/, and/tmp/), with all other paths being read-only or inaccessible. - Network traffic is routed through an OpenShell gateway, which includes an HTTP CONNECT proxy, ensuring agents can only connect to explicitly approved hosts.
- OpenShell's policy enforcement engine provides granular control by evaluating actions at the binary, path, and method levels.
- The system supports live policy updates and maintains a comprehensive audit trail of all 'allow' and 'deny' decisions for oversight.
- NemoClaw incorporates a privacy router that intelligently directs sensitive data to local models, only routing to external frontier models (like Claude or GPT) when permitted by defined cost and privacy policies.
- The NemoClaw command-line interface (CLI) orchestrates the entire stack, including the gateway, sandbox, inference provider, and network policy.
- Hermes Agent employs a skills system for self-evolving capabilities, capturing learned experiences as reusable skills and maintaining persistent memory across sessions in
MEMORY.mdandUSER.mdfiles. - NVIDIA NeMo is a broader, agent-first open suite of libraries that provides tools for accelerating AI agent specialization, optimization, and governance, integrating with various AI tools and agent frameworks.
- NeMo components include NeMo Curator for data preparation, Nemotron for models, NeMo Evaluator for benchmarking, NeMo Retriever for Retrieval Augmented Generation (RAG), and NeMo Guardrails for safety and policy control.
๐ฎ Future ImplicationsAI analysis grounded in cited sources
Enterprise adoption of autonomous AI agents will accelerate significantly.
The provision of robust, kernel-level security and governance frameworks like NemoClaw and OpenShell addresses key enterprise concerns regarding data privacy and system integrity, enabling wider deployment of self-evolving agents in sensitive workflows.
The role of AI security specialists will become increasingly critical in enterprise AI deployments.
The complexity of defining and enforcing granular, dynamic security policies for autonomous agents that can modify their own environment necessitates specialized expertise to configure and monitor systems like OpenShell effectively.
Hybrid AI architectures combining local and cloud-based models will become standard for enterprise agents.
NemoClaw's privacy router, which intelligently directs sensitive data to local models while leveraging cloud models for less sensitive tasks, supports a hybrid approach that balances compliance, performance, and cost.
โณ Timeline
2023-03
NVIDIA NeMo (Neural Modules) framework, focused on conversational AI models, was developed.
2026-02
Nous Research launched Hermes Agent, an open-source autonomous AI agent.
2026-03-16
NVIDIA OpenShell and Cisco AI Defense announced collaboration for securing enterprise agents.
2026-03-17
NVIDIA NemoClaw, an enterprise AI agent framework built on OpenClaw, was introduced.
2026-03-18
NVIDIA OpenShell, an open-source security runtime for AI agent deployments, was detailed.
2026-06-01
NVIDIA announced new software, open-source models, and partnerships, including the availability of NemoClaw and early preview of OpenShell; Canonical integrated OpenShell as a snap for Ubuntu.
๐ Sources (15)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
๐ฐ
Weekly AI Recap
Read this week's curated digest of top AI events โ
๐Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: NVIDIA Developer Blog โ