What an intelligent residence should still do when the internet fails

One-sentence thesis

The most valuable intelligent residence is not the one with the most automation. It is the one whose critical functions remain local, understandable, manually overridable, documented and safe when the cloud, internet or vendor is unavailable.

What this note covers

• critical functions that should remain local;
• deterministic control and optional AI layers;
• manual override and safe failure;
• account ownership and vendor exit;
• bounded AI use cases;
• privacy and staff dignity;
• commissioning during outage, restart and cloud loss;
• what current systems do not reliably understand.

Intelligence should sit above a reliable control spine

Established system practice Evidence-based recommendation

A serious residence separates five layers.

1. Protection layer

Hard safety and protection logic such as electrical protection, leak isolation, fire and life-safety systems, equipment interlocks and critical alarms.

2. Local control layer

Deterministic functions such as essential lighting, basic HVAC, humidity control, shades, pumps, valves, gates and core scenes.

3. Supervisory layer

Local dashboards, schedules, alarms, trend logs and cross-system coordination.

4. Intelligence layer

Prediction, optimisation, recommendations, fault prioritisation and natural-language assistance where the use case is bounded and reversible.

5. Human and operating layer

Owner decisions, staff routines, maintenance, manual override, escalation and documented responsibility.

The design principle is:

AI may recommend, schedule and prioritise. Deterministic logic protects, enforces and fails safe.

What should remain useful without cloud access

Established cybersecurity and resilience principle Evidence-based recommendation

At minimum, the residence should retain local access to:

• essential lighting and basic scenes;
• primary HVAC control;
• humidity and dehumidification control;
• leak detection and defined shutoff logic;
• essential access and gate operation;
• critical technical alarms;
• backup-power and generator logic;
• manual controls for water, air, lighting and access;
• safe restart after power interruption;
• local operating information for staff.

Cloud reporting, remote access and AI recommendations can add value. They should not be the only path to a safe or usable building.

Manual override is part of the design

Established system practice DALLEAU project target

Manual override should be:

• physically available;
• clearly labelled;
• easy to explain;
• protected against accidental misuse where necessary;
• documented in the Building Passport;
• tested during commissioning;
• included in owner and staff training.

A manual control hidden in a technical cabinet, known only to one integrator, is not a credible fallback.

The owner needs custody, not only access

Established governance principle Evidence-based recommendation

Before procurement, the project should define:

• who owns the main accounts;
• who controls administrator rights;
• how staff access differs from owner access;
• how vendors receive temporary access;
• what data can be exported;
• where backups are stored;
• what happens if the integrator changes;
• what happens if a subscription ends;
• which functions depend on a proprietary cloud;
• how firmware, licences and replacement parts will be managed.

A polished app is not an asset-management plan.

Bounded AI use cases can be useful

Growing evidence Pilot-ready under controls

Arrival preparation

Use a confirmed arrival window to prepare protected rooms, water systems and selected services without operating the whole residence at full output for unnecessary periods.

Fault detection and maintenance

Use runtime, temperature, pressure, flow or alarm histories to identify abnormal behaviour and prioritise inspection.

Leak and water response

Use deterministic local shutoff logic for protection, with AI or analytics limited to pattern detection and escalation.

Energy and backup coordination

Use forecasts and load history to recommend or schedule bounded actions while protecting critical circuits through fixed priorities.

Staff workflow support

Turn alarms and asset records into room-readiness checks or work orders without building person-level surveillance histories.

Owner reporting

Summarise system conditions, unresolved faults and maintenance status using structured source data that remains inspectable.

These use cases are credible only when the residence has good sensing, metadata, commissioning and operating discipline. AI cannot repair missing point lists, undocumented valves or bad installation.

Privacy should be designed into the architecture

Established privacy principle Legal-privacy review required

The Atelier method defaults to:

• purpose limitation;
• data minimisation;
• local processing;
• role-based access;
• short retention where detailed data are unnecessary;
• early aggregation;
• owner export and deletion rights;
• visible staff and guest information;
• no hidden behavioural inference.

DALLEAU’s Atelier method excludes:

• microphones as ambient residential monitoring;
• indoor cameras as routine intelligence inputs;
• biometric templates;
• wearables as required project evidence;
• emotion inference;
• health inference;
• person-level behavioural profiling.

A private residence contains owners, family, guests, staff and contractors. One owner approval does not automatically resolve every other person’s privacy and dignity.

Commission failure, not only the ideal scene

Established commissioning principle DALLEAU project method

The system should be tested during:

• loss of internet;
• loss of cloud service;
• local network interruption;
• controller restart;
• power interruption and restoration;
• failed sensor;
• unavailable vendor account;
• manual bypass;
• staff role change;
• owner account transfer;
• backup restoration.

The test should record what continues, what stops, what fails safe, what needs human action and how the system explains its state.

What the evidence did not prove

The evidence does not establish that:

• a residence can understand the family’s intentions in a general human sense;
• AI can safely control every critical function without deterministic limits;
• one interoperability standard solves every integration and lifecycle problem;
• a cloud service will remain available throughout the life of the building;
• an automated recommendation is appropriate for every guest or staff member;
• more sensors produce better decisions;
• cameras, microphones or behavioural inference are necessary for an intelligent residence;
• “future-proof” is a credible guarantee.

Unresolved questions

• Which open protocols and vendors will remain supportable in Bali over ten or more years?
• What level of local spare parts and integrator response is realistic for each system?
• Which AI recommendations create measurable operating value without excessive data collection?
• How should account succession work for family offices, trusts and changing operators?
• Which room-level occupancy signals are useful without becoming person-level surveillance?
• How should system explanations be preserved when different vendors supply different layers?
• What should the residence do when an AI recommendation conflicts with a recent human override?

Sources

The source finding and DALLEAU's residential interpretation are kept separate. Links open the original publisher or authority record.

1. Cyber Security for Consumer Internet of Things: Baseline Requirements

   Authoritative guidance. A consumer-IoT cybersecurity baseline for product and system requirements. Limit: A claim that a residence is unhackable or legally compliant in every jurisdiction.

2. IoT Device Cybersecurity Capability Core Baseline

   Authoritative guidance. A starting set of device cybersecurity capabilities for IoT acquisition and integration. Limit: A complete project threat model, legal advice or a guarantee of security.

3. NIST Privacy Framework: A Tool for Improving Privacy through Enterprise Risk Management

   Authoritative guidance. A voluntary framework for identifying and managing privacy risk. Limit: Jurisdiction-specific legal compliance or a complete residential data-governance plan.

Related content

• Theme: Intelligent residences and local-first control
• Protocol: Intelligent Residence Readiness Review
• Protocol: Building Passport and Data Stewardship Plan
• Methods Note: Why commissioning matters more than handover

Closing

Build intelligence around control, not dependency

DALLEAU Atelier frames use cases, local fallbacks, manual control, data boundaries, commissioning and asset memory before technology procurement hardens.