Describe the capability.
A robot advertises crop scouting. A valve advertises irrigation. The application selects compatible assets from the registry.
AGRICULTURAL INTELLIGENCE OPERATING SYSTEM
Make your devices work together.
A field reading becomes a decision. A decision becomes an approved action.
Connect a device and follow the entire loop.
THE PROBLEM AGRIOS ADDRESSES
Different devices. Disconnected data.
One shared language for useful action.
CONNECTED FIELD / IRRIGATION
Ready to exploreSee the decision in the field.
Drag to orbit · scroll to zoom · select a device
SensingActionNeeds attentionSIMULATED ENVIRONMENT
02 / CHANGE THE INPUTAdjust a condition, then run the loop
LIVE CONTRACT INSPECTOR
Preview01 / Data Acquisition
↳ INPUT
↗ OUTPUT
AUDIT TRAIL
What just happened
i
A working concept, with transparent limits. Inputs and outcomes are simulated; time is accelerated. The advisory uses visible rules and simplified models. This demo does not connect to physical hardware or claim validated agronomic performance.
CONNECT ONCE. REUSE THROUGHOUT THE LOOP.
Device studio
A capability tells AGRIOS what a device can do. An adapter makes its data understandable.
Connected devices
Session registry · these are simulated connections. Select a device to inspect its contract.
Device contract
TRY A REAL INTEGRATION STEP
Different payload. Same meaning.
Edit the raw reading, choose its format and validate. A valid observation updates the irrigation input; invalid data is rejected.
A real deployment supplies an authenticated transport and an adapter for each vendor schema.
NORMALIZED OBSERVATION
Waiting for a payload.
THE ARCHITECTURE, MADE TANGIBLE
Seven layers. One closed loop.
Each layer has an input, a responsibility and an output. Click one to trace the current scenario.
INPUT
OUTPUT
The platform coordinates. The machine controls.
AGRIOS proposes and supervises missions. A real robot must retain local motion control, obstacle response, emergency stop and operating limits.
Approve a bounded action.
Recommendations include evidence, target and quantity. Invalid or stale input prevents a new action; changing inputs invalidates the previous decision.
Check what happened.
Completion includes observed state and an event record. Issuing a command is not proof that a machine performed the task.
What this demonstration implements
| In this browser | Required for a field deployment |
|---|---|
| Simulated registration and common JSON contracts | Authenticated device registry, adapters and event transport |
| Deterministic quality gates and auditable rules | Calibrated data, persistence, permissions and monitored services |
| Water-volume calculation and simplified humidity response | Site-calibrated models, uncertainty and field validation |
| Animated mission, controller acknowledgement and feedback | Hardware integration, local safety control and verified execution |
Model assumptions & equations
Irrigation: gross volume = (target VWC − current VWC) ÷ 100 × area × root depth × 1,000 ÷ application efficiency. This is a uniform soil-water storage illustration, without infiltration, runoff, rain, soil variability or measured crop demand.
Scouting: a predefined field-edge path and synthetic observations demonstrate mission dispatch and feedback. The demo does not perform navigation, crop recognition or obstacle avoidance planning.
Greenhouse: a simplified response toward a humidity target illustrates ventilation control. Actual humidity response depends on outdoor absolute humidity, temperature, crop transpiration and airflow. A high RH threshold here is illustrative, not a disease diagnosis.
AI Advisory layer: the original architecture allows AI services. This demonstrator uses inspectable deterministic rules; no language model or trained predictor is running.
Technical contracts and an implementation path are included in INTEGRATION_GUIDE.md and ARCHITECTURE.md.
NO BLACK BOX
From intent to evidence.
Inspect the latest run: conditions, decision, execution and feedback.