While connected
Existing controllers retain local control. An upper LLM layer analyzes collected data and proposes plans and bounded actions. Only approved actions pass through the control adapter.
Usage example · Smart farm
From control logic to intelligence
Add intelligence above existing farm control
Keep existing local control while an LLM layer analyzes data and proposes plans and bounded actions.
From event to outcome ↓
Connection layout
Sensors, existing local controller, edge buffer, LLM intelligence layer, operator
Trigger
A crop or environmental change, or a network outage
WMOS
From event to outcome
WMOS forwards messages. Adapters implement data interpretation, permissions and business rules.
Existing controllers operate within deterministic control and safety rules
An LLM layer analyzes collected data and proposes bounded actions
Send approved actions through the existing control path and record results
Keep local control and buffering offline, then synchronize after reconnection
While offline
The concept retains rule-based local control and stores history in an edge buffer. A lightweight local LLM is optional and unvalidated. Results synchronize after connectivity returns.
Before implementation
Required adapters
Adapters validate controller commands, permissions and safety limits. The LLM does not replace deterministic control or safety functions.
Approval and constraints
Plans or control changes outside agreed limits require operator approval.
Control from your familiar chat
Control connected systems from your familiar Codex or Cowork chat with a plugin or adapter.
Supported environments require receiver setup, network access and permissions for the target system.