From farmer intent to verified field action
One workspace holds server, website, app and device modules, each with its own README. No module invents its own representation of device IDs, commands, events or states — shared schemas are the contract.
Farmer operations and fleet administration
Mobile-first farmer interface plus a web dashboard for administration, support and analytics.
- Farm and pond creation
- Gateway provisioning and secondary pairing
- Current known state of devices
- Safe ON/OFF command for an aerator/motor group
- Schedules and alerts
- Administration and farm management
- Device/fleet management and support tooling
- Analytics and execution history
- Audit trail for automation and manual control
APIs, auth, device services, telemetry, commands, automation, AI orchestration
Administration and farm-management dashboard
Farmer-facing mobile application
Gateway and secondary-controller firmware
Protocol schemas, types, validation and common constants
Architecture, deployment, electrical integration, LoRa protocol, operations
Device registry, telemetry, commands and audit
APIs, authentication, device services, telemetry, commands, automation and AI orchestration.
| Entity | Key data |
|---|---|
| Farm | Owner, location, configuration, timezone, status |
| Pond | Farm, pond ID, metadata, connected devices, operating profile |
| Device | Device ID, type, firmware, credentials, gateway, last seen, health |
| Motor / Aerator | Pond, controller, control channel, operating state, schedules |
| Command | Target, requested state, issuer, timestamp, sequence, acknowledgement |
| Event | Type, source, timestamp, payload, severity, correlation ID |
| Automation | Trigger, conditions, action, safety constraints, execution history |
| User | Identity, role, farm access, permissions |
Ordered tasks, shared context, deterministic safety
AI agents must follow the correct order, preserve context, work concurrently where appropriate, and execute accurately rather than behaving as isolated chat sessions.
- 01PlannerDecomposes a high-level request into explicit tasks.
- 02Context serviceSupplies relevant farm, pond, device and historical state.
- 03Task schedulerDetermines dependencies and which tasks may execute concurrently.
- 04Workers / agentsPerform bounded tasks with typed inputs and outputs.
- 05Safety / policy layerValidates commands before field devices receive them.
- 06State managerRecords task status, retries, failures and completion.
- 07Audit layerRecords who/what initiated an action and what happened.
Concurrency is dependency-aware: checking several ponds can run concurrently, while a motor command waits for relevant context and safety checks. Conflicting commands targeting the same actuator require deterministic arbitration.
- Every task receives a task ID and correlation ID
- Context is scoped to farm/pond/device, not the whole conversation
- Tool results affecting future execution are persisted
- Critical actions use structured parameters, not free-form text
- Completion is based on verified device acknowledgement where possible
AI and cloud request actions — the field layer decides
Aquaculture field systems operate in wet, electrically noisy and connectivity-variable environments. Reliability must be designed in.
Gateway/node recovers automatically to a known safe state.
Local control and essential schedules continue where configured.
Secondary follows explicit communication-loss policy; no unsafe uncontrolled switching.
Command ID/sequence prevents repeated execution.
Expiry/deadline prevents old commands from being applied later.
Watchdog and rollback/recovery strategy.
Existing industrial protection remains authoritative.
Every remote control action is recorded with source and result.
AI cannot bypass deterministic safety and authorization layers.
Critical control principle: AI and cloud services should request actions; the field control layer decides whether an action is currently safe and valid. Identity, authorization, command validity and physical safety are separate controls.