How it works

How an answer comes together.

You ask, in plain language or through another system, and a sourced answer comes back. It gets there in four steps: connect, understand, control, deliver.

An electrical substation seen from above at dusk.

The pipeline

One path, four steps. 

From the systems you already run, to an answer you can act on.
01

Connect

Reads from the systems your company already runs, through each system's own interface. Nothing gets replaced or migrated.

02

Understand

Works out who is who, what is still true, and where each fact came from. So it never confuses two records for the same thing, or hands back something stale.

03

Control

Before it answers or hands anything off, it checks who is asking, what they may see, and what needs sign-off. Nothing happens outside those limits.

04

Deliver

The same engine answers wherever the work happens. A person's question, a report, or another system asking through an API.

The architecture

Eight layers, one governed record. 

The memory an answer draws on is eight layers deep, from the systems at the bottom to the person at the top. Select any layer to open it.

Integration Layer

Plugs into the systems you already run.

It reads each system through its own connection, over the OT protocols you already use. Nothing is replaced or moved.

OT Protocol GatewayHistorian ConnectorsAsset, GIS & Work APIsEvent BusEdge Collectors

The flow of one agent

Watch one agent walk the path. 

Take the Alarm Intelligence Agent in the middle of an event. It runs the same four steps, then stops where control begins.
1

Observe

It watches the same sourced record everyone else does: the alarm stream, historian tags, the topology, and prior events. Read-only, like the rest of the platform.

2

Reason

It groups the flood, drops the duplicates, and works back to a likely root cause. Every step stays tied to a source, and to the moment it was true.

3

Recommend

It hands the operator one ranked event, a suggested response, the evidence behind it, and a confidence score. A thousand alarms become one story.

4

Hand off

Nothing is actuated. The operator decides. The system that owns the action carries it out, and owns the rollback.

Control begins here

Every agent stops at the same place. It recommends; a person approves; the owning system acts.

The roster

The agents that walk it for you. 

Nine agents, grouped by the work they do. Two are sector-specific, for electricity and gas; the rest serve both. Open any one to see what it reads, what it produces, and where it hands off.

Operations

Operations

Alarm Intelligence Agent

Turns an alarm flood into a few ranked events, each with a likely cause and a suggested response. The operator reads one story instead of a thousand alarms.

Operations

Truck Roll Agent

Decides whether a job really needs a field visit, and when it does, groups and sequences the visits so crews drive less.

Reliability

Reliability

Asset Health Agent

Keeps a live health score for every monitored asset, so wear shows up as a trend rather than a surprise failure.

Reliability

Reliability Agent

Weighs where reliability risk sits, and what each fix is worth, so the riskiest gap gets attention first.

Reliability

Maintenance Agent

Turns a predicted failure into a draft work order, with the parts and crew already lined up, ready for a planner to approve.

Engineering

Engineering · Electricity

Grid Optimization Agent

Simulates the live electricity network and recommends dispatch and switching changes that stay inside reliability and constraint limits.

Engineering · Gas

Gas Network Agent

Simulates the live gas distribution network and recommends pressure and flow setpoints that hold delivery steady within safe limits.

Governance

Governance

Compliance Agent

Keeps regulatory evidence current and audit-ready, so an audit becomes a query rather than a scramble.

Governance

Executive Agent

Pulls operations, finance, and risk into one clear briefing, built on the same facts the control room already sees.

Vector Sheets

The agent that understands your spreadsheets. 

Vector Sheets is a structured-data agent. It reads the tables across your engineering workbooks, answers questions and runs the calculations, and understands how tables on different sheets relate, not just the cells in front of it.

Your numbers live in complex workbooks.

The calculations that run the operation sit in Excel: load and flow models, asset registers, setpoint and rating tables, often many tables spread across many tabs, built by hand over years.

The answer spans sheets, not cells.

The real question usually crosses several tables that reference each other from different sheets. A single formula never captures it.

It answers questions and does the math.

Ask in plain language. Vector Sheets runs the query and the calculation across the workbook, the way a careful engineer would.

It understands the relationships.

It works out how a table on one sheet connects to a table on another, so the numbers line up instead of contradicting each other.

Input

A workbook with many tables across many sheets, however it happens to be laid out.

Read

Finds each table, its headers and totals, and the formulas behind them.

Relate

Works out how the tables on different sheets reference and depend on each other.

Answer

Runs the question or calculation you asked for across the whole workbook.

Output

The answer, with the exact cells and tables it came from attached.

The guarantees

The rules the whole path follows. 

Whatever runs along that path, a question or an agent, holds to the same four rules.

Safe by default

Reading is the default. Anything that could write is explicit, approved, reversible, and logged, with segmentation aligned to IEC 62443 zones.

Nothing unattributed

An answer carries its source, the model that produced it, and a confidence score. Nothing reaches an operator without its evidence.

Grounded in engineering

Statistics stay inside simulation and engineering limits, so a recommendation is one you could actually carry out.

Inside your boundary

Runs in your cloud, your own data center, or a hybrid with inference close to the field. It does not assume public multi-tenancy.

Start a pilot

Run a pilot on your own systems. 

Pick one event that took too long to understand. We stand up a scoped pilot on your own data and reconstruct it, sourced end to end. One use case, read-only, nothing to rip out.