Auri — the edge‑native Home Brain
Auri is Estincelle’s local multi‑agent platform that runs on in‑home hardware, learns household routines, and orchestrates devices across brands with strict latency and reliability guarantees.
What Auri does
Auri runs as a persistent service inside the home. It maintains a live understanding of the home's state and coordinates devices, scenes, and services.
What you get is one Home Brain that reasons about routines, goals, and context, and uses multiple agents to execute the right actions at the right time!
Architecture at a glance
- AI layer - multi‑agent brain
Strategy agents learn long‑term patterns; planner agents turn goals into plans; reflex agents handle sub‑second safety and device actions - World model & orchestration
A shared model of rooms, devices, occupants, and routines with a message bus and event log for learning and analytics - Physical & ecosystem layer
Connectors for existing hubs and protocols (Wi‑Fi, Zigbee, Matter, vendor APIs). Can run alongside or embedded within current stacks.
Designed for real constraints
- Deterministic local behavior under WAN/cloud outages
- Predictable latency for wake words and safety
- Data stays in the home by default
Built as a multi‑agent system
Auri is not a monolithic model. It’s a system of specialized agents with clear responsibilities, shared memory, and measurable SLOs
- Reflex agents -> Fast, deterministic responses for safety and low‑level control
- Planner agents -> Multi‑step planning over seconds with schedules, preferences, and constraints
- Strategy agents -> Longer‑term adaptation across seasons, usage, and energy patterns
- Tools & skills -> Calendar, weather, energy prices, OEM APIs through a controlled interface
Auri — the voice of your home
Auri has an embedded conversational agent. It turns speech, text, and app interactions into structured intents that agents can act on, running as locally as possible (wake‑word, ASR, speaker‑aware behavior)
- Natural language routines (“start my morning”, “I’m going to bed”)
- Speaker‑aware behavior (who is asking)
- Persistent goals and preferences stored in the world model
- Works across voice, app, and physical controls — not voice‑only
Integration & deployment
- Deployment targets — Embedded on hubs/routers; sidecar on existing gateways; companion app on mini‑PC/NAS
- Integrations — Connectors to existing platforms and OEM APIs; optional SDKs for domain‑specific agents
- Pilot flow — Scope → reference deployment → limited field trial → scale‑up with support
Next steps
Auri is in active development and running in internal environments. We’re onboarding a small number of OEM partners
Discuss a pilot
