Context Grammar · One Tuesday with Hana

Same AI. Same person. Five moments.
Watch the interface adapt.

Most AI products treat your context as one thing — busy or not, public or private. Context Grammar reads eight signals at once, plus what your past has taught it, and gets a different answer to a different question every time. Scroll through Hana's day. The story is the demonstration.

Start the day ↓

07:02

Kitchen · stirring breakfast with one hand

Aoi (15) is reheating leftovers while watching Korean drama clips on her phone. Sota (6) is fighting with the cereal box. Hana glances at the fridge — she has 90 seconds before the school run.

What the fridge shows

Tonight: Aoi-safe miso soup
+ leftover salmon

Ready in 18 min · ¥0 added cost

Why one card, not six?
  • Cognitive Load · high — three demands competing for 90 seconds
  • Physical State · one hand — pan in the other
  • Form Factor · fridge display — wet hands, glance distance
  • Brain · Identity Layer — Aoi has a salmon allergy; that's not a preference, that's a hard fact

Six options would be the right answer at 8 PM on Sunday. At 7 AM on a Tuesday, with a child holding a cereal box upside down, six options is cruelty. The framework reads the moment and proposes one.

08:15

In the car · rush hour · Aoi forgot her cleats

Hana is driving Sota to school. Aoi has soccer practice at 4 PM and her cleats are sitting by the front door. The phone buzzes.

What CarPlay does

"Aoi's cleats are at home.
Drop Sota first, then loop back? Or notify Aoi to grab them at lunch?"

Say one or two.

Why no screen?
  • Physical State · driving — hard rule. Visual attention is locked.
  • Form Factor · car display — voice-primary, glance-safe only
  • Priority Weight · high — recoverable but time-pressed
  • Brain · Now Layer — knows Aoi's calendar (4 PM practice)

There is exactly one design rule that is non-negotiable: when the user is driving, there is no UI. The framework refuses to compete for visual attention even when it has good news. Voice + binary choice. That's the whole interface.

12:30

Desk · headphones on · Focus Mode active

Hana is twenty minutes into a design review presentation. Slack pinging, Gmail filling, Calendar trying to reschedule. Notifications all suppressed by Focus Mode. Then the school nurse calls.

What breaks through Focus Mode

Critical · school

Sota — 38.1°C fever
Nurse asks for pickup

14 other notifications stayed silent.

Why this one notification, when 14 others stayed silent?
  • Priority Weight · critical — child safety beats everything
  • Brain · Identity Layer — "school + fever" is in Hana's hard-priority list
  • Cognitive Load · high (focus) — interrupt only if it's worth it

Focus Mode is a vow, not a wall. The framework treats family safety as one of a few things that can override a user's stated "do not disturb." Everything else — Slack, Gmail, the four marketing emails — waits. The 14 muted notifications are the demonstration that this is restraint, not a bug.

18:30

Living room · TV cast · whole family on the sofa

The interesting one

Hana asks the TV: "what should we eat tonight?" — three candidates appear. The kids can see everything. Including prices. A naive system would blur them ("kids around → hide cost"). Context Grammar shows them.

What the TV shows

Salmon poke bowl

25 min · ¥1,820 · Aoi-safe

Tonkatsu set

35 min · ¥2,400

Soba + tempura

22 min · ¥1,520

Why prices are not blurred here
  • Social Exposure · family with children — yes, kids are present
  • Disclosure Dial · full — for the household-meal domain. Brain knows Hana's family discusses budget openly.
  • Intent · joint decision — they're choosing together. Cost is part of the decision.
  • Brain · Learning Layer — Hana has shown prices in this context 47 times. Never asked to hide them.

This is the core argument for Context Grammar over single-rule systems. "Family present → blur" is the kind of well-meaning rule that infantilizes the user. In this room, prices are the whole conversation: "Tonkatsu is ¥2,400, that's a Friday thing — let's do soba." The framework reads the intent and trusts the family to handle their own budget.

Compare: a different evening

Same room, same kids. But Hana is browsing Aoi's birthday gift on the same TV. Now the framework does blur. Why? Intent flipped. "Surprise" is the goal. Same tokens, different intent, opposite UI. The framework's job is to know the difference.

22:30

Bed · phone in hand · planning next month's trip

Lights low. Kids asleep. Hana is loosely browsing for somewhere to take the family in October. She hasn't decided anything. She's just looking.

What the phone surfaces

Three places, three different reasons

  • Hakone2-night ryokan · ¥48k · onsen for the parents · arcade for Sota
  • Setouchi islandsunusual pick · 3 nights · Aoi just photographed Naoshima for class
  • Karuizawalast year favourite · 2 nights · book by Friday for the same cabin

Pinch any one to widen. Long-press to remove forever.

Why three weird picks instead of "top destinations"?
  • Cognitive Load · low — exploration time, not decision time
  • Priority Weight · low — no deadline, browsing tonight
  • Disclosure Dial · full (travel) — Hana shares calendar, budget history, kids' interests for this domain
  • Autonomy Dial · suggest — Hana wants options, not bookings
  • Brain · Learning Layer — knows what Aoi photographed for class last week

Disclosure is the fuel. With it, the AI can be anticipatory — connecting Aoi's school project to Naoshima, remembering last year's cabin, picking a place Sota will actually like. Without disclosure (a stranger AI), the same query gets you "Top 10 family destinations in Japan." That's the cost of privacy you can choose to pay; the framework is honest about both sides.

What you just watched

One framework. Five different right answers.

Same Hana, same AI. The interface morphed five times because five different signals were the dominant ones in five different rooms. No global rule like "blur in public" can do this work. The framework reads the moment, then chooses.

8 Tokens

Read the moment from sensors, calendar, and behaviour. Not 1 signal — 8 at once.

See the tokens →

3 Brain Layers

Identity (who you are) · Learning (what you've shown me) · Now (what's happening this second).

See the Brain →

23 AX Patterns

Named UI moves the framework can fire. The "Approve gate", "Voice override", "Care nudge", and twenty more.

See the patterns →

For builders

A live token console lets you flip any of the 8 tokens and watch the rule engine evaluate all 23 patterns in real time. Useful if you want to dig into the engine itself.

Open the live console →