Une machine qui se souvient, ressent et anticipe.
sharm est un modèle du monde inspiré du cerveau. Treize modules ancrés dans les neurosciences — de l'hippocampe à l'amygdale — fusionnés en une seule architecture qui comprend l'instant que vous vivez.
Today's models read the world.
They don't live in it.
Large language models compress the internet, but they forget your life the moment it happens. Flip the switch to see the same moment through an LLM — and through sharm.
Memory
A hippocampal loop links days into living context.
Affect
Threat, valence and arousal, read as separate axes.
Anticipation
A forward model projects the next moment before you ask.
What we build to solve it: a world model with a brain's architecture — one that separates memories like a hippocampus, weighs them like an amygdala, and projects the near future before you ask.
Une seule boucle : percevoir, se souvenir, ressentir, anticiper
See the moment
sharm encodes each scene the way a visual cortex would — objects, people, motion and place fused into one world state, on-device.
Bind it to memory
A hippocampal loop separates, stores and later recalls episodes — so context from yesterday shapes what matters today.
Appraise how it matters
Amygdalar heads read threat, valence and arousal, giving every moment an emotional weight instead of a flat label.
Predict what's next
A forward model projects the near future, letting sharm notice, prepare and speak before you have to ask.
De la mémoire personnelle à la conscience à l'échelle d'une ville
A companion that remembers your life
sharm notices the people, places and things that recur in your days, then surfaces the right memory at the right moment — 'you met Rama here last spring.'
Grows from gentle recall into an anticipatory copilot that plans errands, spots first-visits and briefs you before meetings.
Situational intelligence for operations
Ego-view cameras stream to sharm's world model, which flags anomalies, tracks assets and narrates what's happening across a site in real time.
Moves from passive monitoring to predictive dispatch — anticipating bottlenecks and routing the right response automatically.
Context-aware public safety
sharm fuses multi-source sensing with privacy-preserving, on-device redaction so responders get grounded situational awareness without raw data leaving the edge.
Extends to city-scale anticipation: early-warning patterns for crowds, infrastructure and disaster response.
Deux couches, un seul cerveau
sharm model v0
The core brain — 13 neuroscience-grounded modules jointly trained into a single cognitive architecture. Here's what that lets it do:
En savoir plussharm sense
The eyes and ears — a privacy-first on-device sensing stack. Here's what it makes possible:
En savoir plusConstruisez avec nous l'avenir de l'intelligence située
Devenez partenaire, lancez un pilote ou rejoignez l'équipe qui donne une mémoire aux machines.