# architecture

VAOS™ architecture

Six layers, one runtime. Local-first. Sensor-aware. Patent-pending innovations at the perception, contract, and action layers.

Scene Contract spec → Try it →
Patent-pending. Three core innovations of VAOS™ — sensor-aware perception, the Scene Contract validation envelope, and validation-gated action runtime — are protected by patents pending in multiple jurisdictions.
# the-stack

Six layers

Each layer is independently extensible. Each layer publishes typed contracts to the next.

👁️

Observe

Camera streams · ISP · raw inference frontend — vaos.sensors

Step 01
🗺️

Structure

Sensor reliability + Scene Contract — pixels become structured scene — vaos.scene

Patent · 02
🧠

Think

Reasoning over scene + memory + site context — vaos.cognition

Step 03
🛡️

Validate

Action gating against scene confidence + sensor reliability — vaos.validate

Patent · 04

Act

Emit to robot, MQTT, REST, dashboard, log — vaos.actions

Step 05
💾

Learn

Outcomes captured into persistent local memory · federated — vaos.memory

Step 06
# data-flow

Data flow per frame

perception → action → memory loopflow
# Per frame, ~30 FPS, <50ms total latency

[observe]     raw frame + ISP metadata + sensor reliability
   ↓
[structure]   detection + Scene Contract { objects, event, metrics }
   ↓
[think]       reason(scene, memory, site_context) → candidate_action
   ↓
[validate]    gate(action, scene.confidence, sensor.reliability)
   ↓
[act]         emit → robot / API / MQTT / dashboard
   ↓
[learn]       memory.append(outcome) → improves future thinking
# patents-pending

Three patent-pending innovations

// PATENT 01

Sensor-aware perception

Continuous evaluation of sensor reliability — calibration, lighting, occlusion, thermal — fed directly into inference confidence.

// PATENT 02

Scene Contract envelope

Validation envelope around structured perception output. Open schema, defended boundary against malformed or untrusted scene data.

// PATENT 03

Validation-gated action

Closed-loop runtime that gates every action against current scene confidence and sensor reliability before execution.

# why-this-stack

Why six layers, not one model

Vision-language models alone don't move robots. Wiring perception to validated action requires explicit boundaries, typed contracts, and a memory of state.

VAOS makes those boundaries first-class: every layer publishes a typed contract, every contract can be unit-tested, every action is auditable.

ConcernStep
Bad sensor inputObserve
Untrusted perceptionStructure
Reasoning & contextThink
Confidence gatingValidate
Outbound actionAct
State / outcome captureLearn

Ready to build on it?

Run the quickstart, fork a plugin, or read the Scene Contract spec.

Quickstart Scene Contract spec