1
00:00:00,000 --> 00:00:05,950
Joe — Nemotron-Powered Edge AI Agent Factory. Built for NVIDIA Agent Challenge 2026.

2
00:00:05,600 --> 00:00:12,050
The problem: vague enterprise requests, heterogeneous fleet, no observability, no delivery.

3
00:00:11,700 --> 00:00:18,150
Joe is a local-first AI workforce — software, video, network, WAN — not a content generator.

4
00:00:17,800 --> 00:00:22,750
Why now: agentic AI, local AI, MoE reasoning, observability — all converging on the edge.

5
00:00:22,400 --> 00:00:29,850
Architecture: Model · Skill · Device routing — a system-level Mixture-of-Experts harness.

6
00:00:29,500 --> 00:00:37,950
Six heterogeneous NVIDIA-class nodes: DGX Spark, RTX 5090, two Jetson-class edge, Arc A770, PC orchestrator.

7
00:00:37,600 --> 00:00:45,550
AI host stack: Copilot CLI, Aider, Claude Code, litellm, OpenRelay, MCP — the programs that drive the cluster.

8
00:00:45,200 --> 00:00:54,150
Twenty-four expert agents — including the NVIDIA Omniverse trio: Isaac Sim, Isaac Lab, OpenUSD.

9
00:00:53,800 --> 00:01:00,250
Two hundred and thirty-four callable skills across Superpowers, AirCore, gstack, and OpenClaw.

10
00:00:59,900 --> 00:01:05,850
Pipeline one: vague feature request becomes a verified deployment artifact.

11
00:01:05,500 --> 00:01:11,450
Pipeline two: a web page becomes an MP4, an SRT, and a thumbnail.

12
00:01:11,100 --> 00:01:16,550
Pipeline three: Joe extends from documents to live infrastructure and WAN load balance.

13
00:01:16,200 --> 00:01:22,150
Forty Iron Laws plus a five-discipline loop — automate, evolve, fix, verify, be honest.

14
00:01:21,800 --> 00:01:27,750
Chatbots reply with advice. Joe ships executable, verified work.

15
00:01:27,400 --> 00:01:34,850
Joe — local-first, observable, deliverable. github.com/applelai001/joe