Spatial capture → walkable world
A phone scan becomes solid geometry in about ten seconds.
Physical intelligence infrastructure2026
The live world layer for physical AI.
SOPHIA gives every operator, agent and machine one persistent, live understanding of a physical site.
Working coreReal buildingBuilt solo
Live · local demonstratorWorld interface / alpha
One physical site, held as live state.04 tracks · 03 sources · 01 world
01 / The missing primitiveContext dies at device boundaries
AI is missing a world, not another model.
Cameras, robots and agents already exist. What they lack is a persistent reality they can all inspect, update and act on.
Cameras see feeds
They do not preserve shared context.
Robots map locally
Each machine rebuilds the same place.
Agents reason in text
They cannot inspect persistent live state.
02 / Working proofNot a concept render
A working core, built from the world up.
Commodity inputs became geometry, shared tracks, memory and policy-gated action. The interface below is the system itself.
Two sensors → one track
Camera and low-cost radar resolve into one live person track.
World → measure → act
Measure a route, reason over the site and trigger a spoken action.
03 / Shared infrastructureOne world in the middle
Sensors change. The world remains.
SOPHIA turns fragmented observations into one continuously usable state layer for people, software and machines.
01
Sensors
Observe
Vision · radar · scans · machines02
SOPHIA world model
Understand
Fuse · resolve · remember · simulate03
World primitives
Hold state
Geometry · state · memory · permissions04
Clients
Act
Humans · agents · robots · systemsSOPHIA OS ALPHA
typed contracts · registry · frame graph · replay
LIVE DEMOworld model · fusion · voice · actions
04 / Wedge → platformUrgent first job, expanding world
Security first
Start where reality already has stakes.
Security is the first deployment because the buyer, budget and operational pain already exist. A live world can be tested against concrete events, permissions and response times.
01
Win one urgent workflow.
02
Install one persistent site world.
03
Let every later client reuse it.
The platform logic
Security funds the first site. Every new machine expands the same world.
05 / One world, many clientsExpand without remapping reality
The site becomes shared infrastructure.
One installed world can support new missions without remapping the site or rebuilding context for every sensor, team or machine.
Emergency response
A live mission world for fire, rescue and disaster response.
Example deployment
Two indoor drones enter a smoke-filled, GPS-denied building. LiDAR, thermal and gas observations build an evidence-backed search state from unknown to mapped to searched, while possible victims, hazards and responder positions remain visible to incident command.
Indoor drones + responders → room / hazard / search state → command + safe routing
Healthcare
Presence, fall events and clinical telemetry in spatial context.
Example deployment
Radar detects a floor-level presence while an authorized wearable reports abnormal telemetry. SOPHIA binds both signals to the person and room, then routes the nearest caregiver with the exact location and evidence.
Radar + authorized device → person / room / vital context → caregiver response
Industrial safety
Live people, machines and permissions inside one operating state.
Example deployment
A person enters a live machine cell. SOPHIA fuses camera, radar and machine state, resolves the restricted-zone event and issues a policy-gated alert or stop.
Person + machine → shared zone state → operator action
Facility operations
Persistent assets, access state and change history across a real site.
Example deployment
A technician asks where the affected pump is and for the safest authorized route. SOPHIA returns the asset, current obstacles, access state and prior observations as one grounded task context.
Asset + live access state → grounded route / context → maintenance action
Robot fleets
Maps, memory and coordination that survive device boundaries.
Example deployment
One AMR finds a blocked aisle. SOPHIA updates the shared world once, so every robot and planning agent inherits the same obstruction and reroutes.
Robot observation → shared obstruction → fleet reroute
Inspection drones
A persistent asset world before, during and after every mission.
Example deployment
An inspection drone finds damage on a roof asset. SOPHIA anchors the finding to the site model, preserves it in memory and gives the next mission the same target for verified change detection.
Aerial scan → asset-linked condition → follow-up mission
End state / Site-level autonomy
A loading bay coordinates itself.
A truck arrives while people, an AMR and an automated door share the same zone. SOPHIA holds occupancy, permissions and intent so the door opens only when the bay is clear and the robot moves only when route and policy agree.
Truck + people + doors + AMR → one operating picture → coordinated action
06 / FounderBuilt close to the field
Leon Karim Valentin · FounderSoftware, hardware and the physical site—built as one system.
05 weeks
From typed contracts to a working live world.
19 years
Inside the first site and close to its operators.
Built the core. Now build the company.
Leon Karim Valentin, 19, built SOPHIA's working core solo in five weeks—across software, hardware and the physical site itself.
The first building is not a staged lab. Leon grew up inside it. That access created a direct loop between real operators, real constraints and the system being built.
Full-timeReal-site accessPre-seed
For investors, design partners and builders