It fits the SCADA you already have
The question everyone asks: “does this force me to run another system next to the SCADA?” No. The assistant consumes the interfaces the platform already publishes. Nothing more.
On your SCADA
The SCADA remains the source of truth and the operator workstation. Siemens, AVEVA / Schneider Electric and Rockwell Automation platforms already publish what Wabby needs: OPC UA, SQL for history and alarms, REST or GraphQL APIs, MQTT with Sparkplug B, scripting and web screens embeddable in the HMI. Wabby reads through those doors and returns the result inside the HMI.
Rules of the fit
- 01The conversational assistant reads a lot and writes little; when it writes, it is under human approval and audited.
- 02Automated control and optimisation remain deterministic models, validated and executed on the platform.
- 03No model inside the functional safety chain.
- 04Each vendor's engineering assistants help build the project; Wabby helps operate it day to day.
Siemens, AVEVA / Schneider and Rockwell
We have programmed and maintained these platforms at customer plants. Wabby connects to each one through its standard interfaces, with no proprietary modules.
Siemens
WinCC · WinCC Unified · WinCC OAOPC UA, process and alarm archives in SQL, web screens in the HMI.
AVEVA · Schneider Electric
System Platform · InTouch · PI SystemOPC UA, Historian via SQL, PI Web API (REST).
Rockwell Automation
FactoryTalk View · FactoryTalk Optix · FactoryTalk HistorianOPC UA, history via SQL, MQTT.
Who does what: the platform controls and runs its numerical models; Wabby runs the language model and guides people.
And with any other platform
Wabby doesn't come in through the SCADA brand but through what it publishes: OPC UA, SQL, REST, MQTT. If your platform exposes it, Wabby reads it. If not, the first thing we do is expose it. That's our trade.
Where it lives
The same Wabby in three deployments. The lock shows where the data lives.
Edge
A server next to the SCADA, small model, no internet egress. For plants that want nothing to leave.
On-prem
Plant server with GPU, mid-size model. The usual balance.
Hybrid
Index and permissions at home; model via API with anonymised data. Only if your policy allows it.