sharm
Product

The brain, and the senses that feed it

sharm ships as two layers — a world model and a perception stack — trained and designed to work as one cognitive system.

A moment with sharm

Not features on a list — a mind living the moment with you

sharm's point of view in a café
LIVE · sharm sense
person
laptop
plant
cup
Recall

Blue Bottle · you were here Tuesday with Mei

Threat
Valence
Arousal

Your 3:00 is 10 minutes away — want to head out?

It sees the whole scene

Objects, people and place — fused into one world state, on-device.

In development

The device that notices — revealed soon

We're building a single, always-on sensor you simply wear. We're not ready to show it in full just yet — but it changes what a device can notice about your world.

Something you'll want to wear.

Hold tight — the full reveal is coming.

Unveiling 2026
Always-onFeatherlightPrivate by designYou wear it

Want to be first to see it?

Get early access
Coming soonThe world model

sharm model v0

The core brain — 13 neuroscience-grounded modules jointly trained into a single cognitive architecture. Here's what that lets it do:

6 capabilities

Remember your days

Recall the people, places and moments from last week — or last year — so context carries forward.

Hippocampal memory loop

Feel what matters

Weigh each moment by threat, valence and arousal — it can tell quiet caution from real overwhelm.

Affect heads · r ≈ 0.87

Anticipate the next moment

Project the near future and prepare — noticing before you have to ask.

Multi-horizon prediction

Speak up — only when it counts

Decide what's worth mentioning and surface it, staying quiet the rest of the time.

Selective surfacing gate

Run live on your device

The full perceive–remember–feel–act loop, in real time, on the edge.

7.3M-param unified brain

Know when it's unsure

Flag its own low confidence and hand off to a specialist instead of guessing.

ACC uncertainty + handoff
Coming soonThe perception layer

sharm sense

The eyes and ears — a privacy-first on-device sensing stack. Here's what it makes possible:

6 capabilities

See and name your scene

Detect, recognize and narrate the objects, people and places around you, on-device.

On-device detection + narration

Learn from your words

Teach it something once — “that's my Yeti mug” — and it recognizes it from then on.

Open-vocabulary learning

Keep raw moments private

Frames are redacted on-device; the raw signal never crosses the network.

On-device redaction moat

Follow conversations

Turn speech and speaker turns into spoken context alongside what it sees.

ASR + speaker tracking

Scale up when needed

Offload heavy perception to the cloud seamlessly when a task needs more compute.

Hybrid edge/cloud offload

Ground every answer

Tie replies to what it actually observed — grounded facts, not hallucinations.

Grounded world state
Architecture

Thirteen modules, one WorldState bus

Every module reads and writes a single shared world state, in a fixed execution order — from the Cosmos visual encoder through the hippocampal memory loop, the amygdalar affect heads, and out to action selection and reflection.

13 modules · execution order
01
Cosmos
Visual world encoder
02
Thalamic
Sensory relay & gating
03
Grid
Spatial coordinates
04
Dentate Gyrus
Pattern separation
05
CA3
Associative memory
06
CA1
World prediction
07
Cerebellar
Fine motor & timing
08
Fast Amygdala
Threat reflex
09
Slow Amygdala
Affective appraisal
10
ACC
Conflict monitoring
11
Self-Model
Sense of self
12
Basal Ganglia
Action selection
13
Default Mode
Reflection & replay
Use cases

Built for consumer, business and government

Consumer

A companion that remembers your life

Wearable memory

B2B

Situational intelligence for operations

Field & floor awareness

Government

Context-aware public safety

Civic & emergency response

Deep-dive each industry