NemoClaw in MSI EdgeXpert-0415

1.

NemoClaw
with MSI EdgeXpert
MSI IPC |Apr. 14th 2026

2.

Agenda
‒What is Agentic AI
GenAI vs Agentic AI
OpenClaw-1st groundbreaking AI Agent
OpenClaw vs NemoClaw
OpenShell
‒Why run Agentic AI in Local
Cost estimation of Cloud AI: Claude & Gemini

3.

GTC 2026 Build-a-Claw Event by Nvidia and MSI

4.

Agentic AI is coming
Image source:
Nvidia CES 2025 Keynote

5.

6.

Source

7.

The Paradigm Shift
Generative AI
OpenClaw (AI Agent)
Reactive Architecture Operates in a passive,
Autonomous Loop Operates in a goal-oriented
"request-response" mode. You provide a
mode. It breaks high-level objectives into sub-
prompt, it generates content, and the task
tasks and initiates a continuous "Thought →
ends. It does not decide what to do next.
Action → Observation" loop until achieved.
Strict Action Boundaries Boundaries are
confined to the "chatbox." Output is limited to
Tool Use & Execution Possesses actuation
generating text, code, or images. It cannot
capabilities. Executes terminal scripts,
physically alter computer states or external
navigates web browsers, reads/writes local
environments.
files, and operates software interfaces as a
digital worker.

8.

Environmental Perception
Debugging vs. Self-Reflection
Generative AI:
Lacks awareness of the actual execution environment. If
generated code contains a bug, it relies entirely on a human
user to copy the error log, paste it back into the chatbox, and
prompt it for a fix.
OpenClaw Agent:
Features advanced self-correction capabilities. It natively
"reads" terminal outputs, incorporates error logs as context,
analyzes root causes, and autonomously rewrites and
executes code without human intervention.
The Bottom Line
Generative AI tells you "how something should be
done," whereas OpenClaw "steps in and actually
gets the job done for you."
Image Source: oliverpickup.com

9.

10.

11.

12.

NemoClaw: Nvidia’s Secure, Enterprise-Grade AI Agent
OpenClaw
Unbound & Agile
NemoClaw
Governed &
Fortified

13.

Nvidia NemoClaw & OpenShell
• NemoClaw: Run OpenClaw more securely
inside OpenShell with managed inference
–
OpenShell
OpenShell is the safe, private runtime for autonomous AI
agents. It provides sandboxed execution environments
that protect your data, credentials, and infrastructure

14.

15.

Nvidia NemoClaw now available
• Safer AI Agents & Assistants with OpenClaw | NVIDIA NemoClaw

16.

NemoClaw’s Architecture
• OpenShell is the sandbox for NemoClaw/OpenClaw
OpenShell
Image Credit: Nvidia

17.

18.

19.

20.

Why run AI Agent in local?

21.

Expense on Cloud AI Model
• Person AI Assistant

22.

Cost Estimation with Cloud AI
A small team of 10 users, each using OpenClaw daily with the following activity profile:
Activity
Volume
Details
Email processing
~100 emails/day
Scanned and analyzed for classification, summarization,
and prioritization
Conversational interaction
30 minutes/day
Active multi-turn chat with the assistant
Text messaging input
~500 English words/day
Text conversations entered via messaging apps
(WhatsApp, Telegram, Slack, etc.)
Web research
150 web pages/day
Pages fetched for content analysis and synthesis
Background processes
Continuous (scheduled)
Memory consolidation and Dreaming (background
reflection/integration)
Tier
Claude
Gemini
Handles