Technical credibility is part of the product, not a footnote.

Home Assistant, the global foundation

Over 2 million active installations, the most-contributed home-automation project on GitHub, and 3000+ integrations. casasmooth generates native Home Assistant configurations: nothing exotic, everything stays standard, readable and reversible.

A dedicated system, not an app

The box runs a dedicated Linux system where every service lives in its own isolated container — the casasmooth engine included. The cloud is decoupled by design: supervision, updates and billing, never in your home's real-time loop.

Local-first, verifiable by pulling the cable

Automations, energy control and security work without the internet — pull the network cable and the equipment keeps being driven. The core runs inside the building. What leaves it for casasmooth is limited to the box's health status and diagnostics, requests to the assistant (hosted language models), remote access, the optional backup and, if you subscribe to them, the telemetry subscriptions. The box also queries third-party services: weather and solar forecasts (with the site's coordinates), electricity tariffs and, for devices that have one, their manufacturer's cloud. The privacy policy details it.

A generated configuration, not a hand-crafted one

The casasmooth engine reads the home's real inventory, categorises every device into a semantic model, then generates automations, dashboards and documentation — reproducibly. A daily quality gate checks model integrity and automation failure rates, and every engine change is validated by comparison before rollout.

Zigbee, Z-Wave, ONVIF and 3000+ Home Assistant integrations

All the protocol compatibility inherited from Home Assistant: Zigbee, Z-Wave, ONVIF cameras, Wi-Fi, Thread — plus the thousands of integrations of its ecosystem. A device Home Assistant supports is a device casasmooth can categorise and drive.

Matter — the home exposed to Google Home

casasmooth exposes its entities as standard Matter devices (lights, plugs, covers, thermostats, sensors): Google Home and other Matter controllers see and control the home, without going through the casasmooth cloud. In the other direction, an off-the-shelf Matter device is added to the home through Home Assistant's Matter integration. The bridge currently identifies itself with a test vendor ID, so a controller may show it as an uncertified device.

KNX — ETS project import

Upload the .knxproj file (password-protected projects supported): ETS Functions yield high-confidence entity proposals, heuristics cover the rest, every proposal explains its score, and doubtful cases require a human review before anything is created.

Loxone — installation takeover

The Miniserver is auto-discovered on the local network, every Loxone control becomes a controllable entity, Loxone rooms become zones — and a background watch re-syncs whenever the installer edits the project. Strictly LAN, never through the Loxone cloud.

Modbus and SunSpec — discovery and adoption

A strictly read-only scan finds meters and inverters on the network, recognises published profiles, generates a complete profile from self-describing SunSpec devices and surveys unknown registers. Adoption is an explicit gesture that always starts disabled: nothing discovered ever acts on its own.

Bluetooth, LoRaWAN and MQTT mesh

Room-level presence by Bluetooth trilateration (phones, watches, beacons — via ESPHome proxies), long-range LoRaWAN sensors received through The Things Network (bridge in service in the cloud, exercised with simulated sensors, no real sensor connected yet), and an MQTT mesh coordinating several buildings or dwellings with each other.

Your needs are contracts

Fixed hours per device (turn on at / turn off at, honoured from the very first minute), charging sessions with an energy target, mandatory daily running hours, minimum temperatures with an anti-legionella floor, only-when-home rules, do-not-interrupt thresholds: the optimizer must plan around them — never the other way round.

A manual gesture always wins

Wall switch, app or dashboard: your gesture takes over immediately. Automation steps aside for a configurable time, then gently resumes — without ever undoing what you decided.

Intelligence in the remaining freedom

24-hour planning of the heat pump, water heater, EV charging and battery: live spot prices and official tariffs, forecast solar surplus preferred at equal cost, re-planning every 30 minutes without undoing what already ran — the energy promised is the energy delivered.

Every decision is auditable

The EMS journal narrates in plain language what was switched on, when, why — and the savings actually measured. Grid-operator signals enter the plan and the rules, acknowledged and journalled.

SmartGridready, from level L1 to level L5

Manufacturer-independent control of SGr (Swiss standard)-labelled devices, heat pumps, chargers, meters and inverters, using profiles from the official library (control validated on a test bench). L5 flexibility covered at 90%, tested with the DSO simulator. Our SmartGridready module for Home Assistant is published as open source. Outside Switzerland, grid control follows each market's own standard: EEBUS (IEC 63380) for §14a in Germany and Austria — casasmooth has written an EEBUS stack, tested on our bench against the reference implementation, not certified and not shipped in the box. Modbus/SunSpec device control is written and validated on a test bench, not yet on a site.

Virtual SGr, the existing building stock counts too

The real flexibility reservoir is not in new builds but in the existing stock where almost nothing carries a label. Any device controllable from Home Assistant, a plain relay or a thermostat, gets the same energy logic without a native SGr interface. A relay costing a few tens of francs is enough to benefit from spot pricing and solar surplus.

Grid operator signals (GRD / VNB / DSO)

Power limits, blocking windows and load management events arrive through a secured and audited interface, then feed into the plan and the dispatch rules. Their physical effect has not yet been measured on a site. Every signal is acknowledged, journalled and reversible, with automatic arbitration when several signals arrive at once, and a thirty-day compliance history.

Self-consumption and local communities

For a self-consumption grouping (RCP / ZEV) behind a single connection point, casasmooth controls the equipment and handles the building's internal allocation. For a local electricity community (CEL / LEG), several homes discover one another, propose collective actions, vote and each execute locally, leaving a verifiable trail. Coordination does not depend on the chosen regime, the same mechanism serves a virtual grouping between neighbouring connection points (vRCP / vZEV). Every home stays sovereign and measurements remain exportable to the grid operator's settlement.

Community coordination is at pilot stage, validated on a multi-instance test bench, not yet offered commercially.

Custom sensors, on ordinary cameras

An AI analyses the frame of any plain IP camera and turns it into arbitrary semantic sensors: is the driveway snowy, is the dog in its kennel, is the parking spot free. No smart camera and no AI hardware required — the CHF 40 camera is what becomes intelligent.

Real-time object detection

Local detection of people, vehicles and animals on the cameras (Frigate NVR), with structured clip archiving and configurable retention. Every detection can trigger a casasmooth automation — lighting, alert, recording.

An assistant that knows your home — and does not make it up

Local speech recognition and synthesis (Whisper, Piper), a multi-provider LLM engine with automatic failover, and explicit anti-hallucination rules: the assistant is bounded to the building's real configuration and checks an automation's actual state before talking about it. It also creates automations from natural language — draft first, explicit confirmation before anything activates.

Advice and vigilance, grounded in your data

Monthly AI-generated reports (energy with real vs potential cost, per-zone air quality, security), a behavioural recommendations engine, and anomaly detection over a 9-day baseline — down to spotting a consumer still following a schedule nobody applies any more.