Accentio EXA Home, Kiosko
Summary
EXA Home, Kiosko is a personal DIY edge AI project (Raspberry Pi + touchscreen + Flutter Pi UI + 3D-printed enclosure designed in Fusion 360) that replaces Alexa / Google Home with a self-owned home automation assistant. Doubles as end-to-end validation of EXA as the agentic AI brain, integrated with Home Assistant, custom MCP servers for container self-management, multi-agent orchestration, and Notion as the human-in-the-loop surface.
Highlights
Self-owned Alexa replacementFull DIY edge AI stack with no cloud activation dependency or vendor lock-in
Replaces Alexa / Google Home dependency with a self-owned voice + touchscreen assistant by combining wake-word detection, voice pipeline, LLM-with-tools, Home Assistant integration, and a 3D-printed enclosure on Raspberry Pi.
EXA validated on edge hardware1 personal kiosk validates EXA platform for real residential use
Validates EXA's full agentic stack end-to-end by running it on edge hardware where EXA serves as the brain and personality for the home assistant.
Agent that manages its containerAgent manages its own Docker container without human intervention
Lets the assistant self-manage its own runtime on the Pi through custom MCP servers for container ops (logs, restart, deploy, healthcheck) via OpenClaw.
Edge and cloud LLM routingMixed edge + cloud routing balances cost, latency and reasoning quality
Minimizes cost and latency through intelligent LLM routing: small / edge models for light tasks, cloud LLMs (Anthropic Claude, OpenAI) for reasoning-heavy work.
Sub-agent coordination1 orchestration pattern supports parallel sub-agent execution and recovery
Coordinates sub-agents through a multi-agent system (parent + sub-agents, shared memory with task graph, message channels).
Notion mission-control boardNotion serves as live mission-control board for the agent system
Gives humans real-time visibility and control through Notion API as the human-in-the-loop surface (live backlog, audit, manual task assignment).
Fully local activationAlways-on listening with zero cloud activation latency
Keeps activation fully local through trained edge ML models (wake word + VAD) running on the Pi without cloud roundtrip.
24/7 unattended operation24/7 reliable residential operation
Operates 24/7 unattended through systemd services, watchdog supervision and proactive MQTT messaging.
Enclosure designed and 3D-printedEnd-to-end DIY: industrial design + 3D print + edge AI software stack
Co-built hardware and software: custom enclosure designed in Fusion 360 and FDM-printed, paired with Flutter Pi UI on Linux headless.