LUNA 2 0 Multi Agent Edge AI with a Local Decision Engine
summary
LUNA is a working Raspberry Pi + tablet prototype that runs a full edge multi‑agent stack, using the OpenAI Responses API, MCP, LangGraph configs and Azure for persistence and telemetry. It wires wakeword→ASR→LLM→TTS in a serial, event‑driven pipeline to avoid parallel resource spikes (Whisper ~40% CPU, Piper ~8%), enforces a 400 ms silence lockout, stores local context (about 122 user attributes), runs planner, emergency, resource and orchestrator agents, and falls back to cloud for heavy reasoning — the demo videos and cost surprises show it actually works.
AI-written summary · about
chapters (7)
- 0:00Luna multi-agent edge AI intro
- 3:40Agentic architecture
- 13:00Real-world demo
- 22:15Architecture principles
- 31:52Agentic patterns and tooling
- 43:17Edge deployment
- 50:23Lessons and future
related talks
30mLook ma, no hands! Multimodal AI-Agents in the browser
16mUnleash the superpowers of the AI PC
47mOwn Your LLM: Running Generative and Agentic AI Locally
22mAI Agents for Multiprovider Cloud Edge Continuum
50mLet's build an AI agent
22mClient side Web AI Agents: Building smarter user experiences for future agentic ...
30mOrchestrating Real-Time Multimodal AI Agents with Rust
25mFrom Prompt to Production: Building Autonomous AI Agents in Node.js