Skip to content

TwinEdge / Platform · Your own assistantPlatform overview

Bring your own AI assistant

Give your assistant operating context.

Connect enterprise AI tools to TwinEdge asset, history, and work information through MCP. Keep source context attached to the answer.

Request a demoThe interface itself

Context that reaches your applications01 / SYSTEM VIEW
EquipmentHistoryDocuments
TwinEdgeShared asset context
Digital twinsAI agentsApps + APIs
01Define02Connect03Authorize04Consume
The application receives a clear contract and traceable source context.
What it can and cannot do
CanCannot
Look up a machine and its historyStart, stop, or adjust anything
Explain what the readings showApprove its own suggestion
Draft a job for reviewClose a job
See one company’s equipmentSee another company’s anything

§ 01

The six rules it runs under.

These are enforced in the software, not requested in a prompt.

  • 01Two kinds of tool, and no third

    One kind looks things up. The other puts a draft into the approval queue. Nothing starts a machine, moves a setpoint, or closes a job.

  • 02People approve

    Every draft waits for a person inside TwinEdge, exactly as it does now. An assistant cannot approve its own suggestion.

  • 03One key per company, per assistant

    Each connection gets its own key and can only see that company’s equipment. The filtering happens in the database, not in the wording of a prompt.

  • 04Everything is written down

    Who asked, which tool ran, what came back, what was proposed, and who approved it. The same record as the rest of the system.

  • 05Answers carry their evidence

    Readings, limits, history, documents. If the evidence is thin, the answer says so rather than sounding equally confident.

  • 06The least data, not the most

    Tools hand back summaries and time windows rather than whole tables. An assistant gets what the question needs and nothing else.

“A rule an assistant can be talked out of is not a rule. It has to be in the database.”

Why the wording of a prompt is not a control

§ 02

Three pieces, depending on which way you need it.

  • 01For your assistant to use TwinEdge

    Ask your own company chatbot about a machine and get an answer grounded in the equipment record, the live readings, and the work history — not in whatever it read on the internet.

  • 02For TwinEdge to use your systems

    The other direction. TwinEdge can read your existing maintenance system, document store, or historian in place. Your data stays where it is.

  • 03For the box at the site

    A small version running on the site box itself, working with no internet at all, so an assistant on a technician’s laptop can ask the machine directly.

§ 03

What people actually ask it.

The useful questions are dull ones: why is this blower running hot, when was this pump last serviced, which machines have degraded most this quarter, and what did we do the last time this happened.

They are dull because they are the questions that currently take somebody twenty minutes and three systems to answer. That is exactly where the time goes.

Point the assistant you already use at one site.

One scoped key, one site, read-only. Let your team ask it the questions they currently walk across the office to ask each other, and see whether the answers hold up.

Request a demoHow approval works