One agent that has read everything the others wrote.
The agents at the line each see one thing well. This one reads all of their records together, alongside your MES and CMMS, and answers the questions that span them — why the shift missed target, whether three stoppages were really one cause.
What it catches
- A quality escape traced back through the inspection records that preceded it
- Repeat stoppages that share a root cause across shifts or lines
- NCR and CAPA drafts assembled from the evidence already retained
- Downtime and OEE explained against what the edge agents recorded
- Audit questions answered from records rather than reconstruction
Where it reports
- MES and ERP quality records through governed HTTPS APIs
- CMMS work orders with the originating evidence attached
- Quality management systems for NCR and CAPA workflows
- Shift reporting and existing dashboards via outbound webhooks
- A retained record naming every source it drew on
What actually runs, and where
- Task
- Reasoning over retained records and systems of record, not sensor streams
- Typical target
- Runs in the control plane; no edge device required
- Mode
- On question, plus scheduled shift and period summaries
- Reach
- Whatever you grant. Commonly read across, write to work orders
Every one of these passes the Oddy test against your named device before it is admitted. A candidate that does not fit the device envelope never receives an edge directive.
Common questions
Does this need the edge agents to be useful?
It is far more useful with them, because their retained records are the evidence it reasons over. It will still work against your MES, CMMS and quality system alone — but the answers get much sharper when it can point at the frame that caused the reject.
Does it run on our hardware?
It does not need to. This one reasons over records rather than sensor streams, so it runs in the control plane. The hardware question applies to the agents at the line, not to this one.
Can it close a work order or sign off an NCR?
It can draft and it can open, where you grant it that reach. Approving, signing and closing are decisions with names attached to them, and we would push back on any deployment that tried to hand those to an agent. What it removes is the assembly work, not the accountability.
How does it avoid inventing an explanation?
Every claim it makes cites the record it came from, and the records are retained independently of the agent. If it cannot point at a source, you should treat the answer as a hypothesis — which is exactly how the interface presents it.
Ask it something you cannot currently answer.
Bring a question your last shift review could not settle. The free audit shows what it would take to answer it from records rather than recollection.