The social-battery pattern: ambient computing that respects human state

What the social battery actually models

The social battery is a numeric state, scaled zero to one hundred, recomputed every few minutes. It rises and falls based on three kinds of signal.

Message-volume drain. Every inbound message — across the messaging stack, email, calendar invites, voice calls — contributes a small drain to the battery. The drain scales with intensity: a casual message is a few units, a long email thread is more, a scheduled meeting block is more again. The exact weights are tuned per occupant.

Sentiment-based drain. Inbound messages run through a small local sentiment model, and the emotional intensity of the message contributes additional drain or replenishment. A warm message from a partner is a small replenishment. A demanding message from a difficult client is a larger drain. The model is local for privacy reasons; nothing about correspondence content leaves the house.

Replenishment cycles. Sleep replenishes. Time spent in the house with no inbound traffic replenishes. A walk, inferred from phone location, replenishes. The replenishment rate is also tunable.

The aggregate is a single number that the rest of the home can read.

What the home does with the number

The behaviour change is subtle and that is the point. We do not want the home to be a personality. We want it to be quietly responsive.

• When the battery drops below a threshold, the alerting layer downgrades all non-critical notifications to digest mode. The phone stops buzzing for low-priority messages. Important alerts still get through.
• The lighting shifts to a warmer, lower-intensity scene. Not dramatically — just enough to register as ambient calm.
• The music agent, if anything is playing, transitions to lower-tempo material from the same artist or genre.
• The calendar surfaces a recommendation to defer non-essential meetings booked for the next two hours, with one-tap deferral available.
• The household briefing agent adjusts its tone — fewer prompts, more compact summaries, no enthusiastic exclamation marks.

None of these are dramatic. Each is a small tilt in the direction of recovery. The cumulative effect is a household that feels noticeably less demanding when the occupants are already drained, and that disappears entirely when they are not.

Why the local-first architecture matters here

The social battery would be unthinkable on a cloud-routed smart home. Every message would have to be sentiment-classified by a third party. The location data would have to live in a vendor's analytics. The aggregate signal would be visible to whichever cloud service was running the orchestration. The privacy profile would be unacceptable.

On a local-first architecture, every component runs on hardware in the house. The sentiment classifier is a small open-weights model on the workhorse. The message volume counter is a script in the orchestration layer. The location signal stays on the phone and is shared with the home only as a coarse zone. The aggregate is computed locally and never leaves.

The architectural commitment is what makes the personal modelling tolerable. Without it, the social battery would feel surveillant. With it, the home is just paying attention to its own signals about its own occupants, and that is what an operator-grade home should do.

Calibrating without becoming intrusive

The hardest part of building the pattern is the calibration. The weights are wrong on the first version. The thresholds are wrong on the first version. The behaviours are too aggressive or not aggressive enough, and tuning them is an iterative exercise over weeks.

The discipline that helps is to log every state change with a timestamp and the trigger that caused it, and to review the log periodically against the occupants' subjective sense of how the day went. If the battery dropped at a moment that did not feel draining, the weights need adjusting. If a behaviour fired that felt wrong, the threshold or the mapping needs adjusting. The log is the source of truth for the calibration.

The other discipline is restraint on the behaviour side. The home should under-react rather than over-react. A subtle dim is better than a dramatic shift. A digest is better than a silence. The pattern is about ambient responsiveness, not about the home asserting itself. Loud responses get turned off; subtle ones become invisible infrastructure.

Generalising the pattern

The social battery is one instance of a more general pattern: infer a slow-moving state about the occupants from observable signals, expose it as a number, let other systems read the number and respond appropriately. The same shape works for several other states.

• Focus state: a number that rises during deep-work signals (calendar focus blocks, do-not-disturb mode, certain music profiles) and falls during context switches. Lighting and notification behaviour respond.
• Recovery state: a number that combines sleep quality, heart-rate variability, and subjective check-in into an aggregate. Heating, lighting, and the briefing agent respond.
• Hosting state: a number that goes high when a guest is in the house, inferring this from device presence and occupant location, and adjusting privacy-sensitive automations accordingly.

Each is the same architectural shape. The work is in choosing the right signals, calibrating the weights, and building the responses with restraint. The compounding payoff is a household that is responsive without being intrusive, and that gets better the longer it runs.

The takeaway

The social battery is the most useful piece of household automation we have shipped, and the one that consumer smart-home products will not build because it requires modelling the occupant in a way that is only acceptable when the modelling is fully local. It is a small architectural pattern, calibrated over time, that makes the home noticeably easier to live in.

If the smart home you have today feels like a collection of clever individual devices, the cognitive domicile is what happens when you give them a shared model of the people inside the house. The pattern is the bridge.