Skip to main content
Import Intel

How it works

From machine signal to reviewed action.

The platform workflow, in the order it actually runs. Nothing in a later stage runs before the stage below it is in place.

The four layers

Connect, understand, govern, execute

01

Connect — Connect the operation

Bring in approved plant, business, document, and market sources — without opening an inbound path into plant systems.

  • Machine, PLC, sensor, and optimizer data
  • Production SQL sources and read-only operational databases
  • Downtime systems and machine-event sources
  • ERP and accounting adapter contracts
  • Work orders, maintenance records, and parts inventory
  • Files, PDFs, spreadsheets, and daily reports
  • Microsoft 365 sources when delegated access is configured

02

Understand — Build operational context

Map machines, metrics, parts, work orders, customers, and workflows into one operational model so employee language resolves to the right facts.

  • Site, process area, machine, line, and source
  • Metric, unit, shift, product, and grade
  • Production events, downtime events, and operator responses
  • Parts, stock balances, locations, assets, and work orders
  • Customers, orders, invoices, and documents
  • Evidence and lineage across all of it

03

Govern — Enforce the boundaries

Agents interpret intent. The platform decides which data, tools, queries, and actions are allowed.

  • Tenant scope and permission boundaries
  • Allowed schemas, tables, and columns
  • Read-only SQL with mutation blocking and statement timeouts
  • Bounded result sizes and supported visualization shapes
  • Source and freshness disclosure
  • Tool risk policies and approval requirements

04

Execute — Deliver useful operational work

Answers, dashboards, reports, and review-ready actions a non-developer can use.

  • Grounded conversational answers and editable dashboards
  • Downtime-reason prompts and incident timelines
  • Ranked action plans with owners and verification metrics
  • Reports, spreadsheet-ready exports, and PDFs
  • Draft invoices, draft purchase orders, and unsent messages
  • Reusable saved views, agents, and skills

Stage by stage

What happens between the sensor and the screen

  1. 01

    Connect at the edge

    The edge collector captures local plant data, queues it durably, retries uploads, and sends it to the platform. It does not require a public inbound connection into plant systems.

  2. 02

    Keep the raw record first

    Source data lands in raw form before anything is derived from it, so a fact can always be traced back to what the source actually reported.

  3. 03

    Extract facts deterministically

    Production, downtime, and event facts are extracted by explicit, idempotent rules — not inferred by a model. The same input produces the same fact.

  4. 04

    Model the operation

    The ontology and semantic layer map employee language to the right machines, metrics, units, shifts, parts, work orders, customers, and documents for that site.

  5. 05

    Plan with agents and skills

    Agents are reusable roles with defined tools, sources, permissions, and instructions. Skills carry the repeatable procedure — the digital version of a standard operating procedure.

  6. 06

    Enforce at the tool boundary

    The model interprets intent; the application enforces tenant scope, allowed schemas, read-only SQL, bounded results, supported visualization shapes, and tool risk policies.

  7. 07

    Produce artifacts and gate actions

    Dashboards, reports, PDFs, drafts, and messages are produced with provenance attached. Anything consequential waits for a person to approve it.

  8. 08

    Keep the audit trail

    Queries, returned rows, time scope, agent activity, artifact identity, and freshness status stay available for review after the fact.

Architecture

  1. 01

    Plant, mill, and office systems

    PLCs, sensors, optimizers, production databases, ERP, documents, email

  2. 02

    Import Intel Edge and connectors

    Local collection, durable queueing, retried uploads — no inbound connection into plant systems

  3. 03

    Facts, ontology, and governed knowledge

    Raw-first ingest, deterministic fact extraction, shared operational model, saved knowledge

  4. 04

    Agents, skills, and tools

    Reusable roles, digital standard operating procedures, governed tool access

  5. 05

    Answers, dashboards, reports, and draft actions

    With source evidence, time scope, and approval gates attached

Provenance

What stays attached to an answer

Results can preserve the evidence behind them, so a number in a meeting can be checked instead of argued about.
  • Source system
  • Source table
  • Source record ID
  • Tenant
  • Source or asset key
  • Event timestamp
  • Returned rows
  • Generated query
  • Calculation steps
  • Artifact ID
  • Model and agent activity
  • Freshness status

First engagement

How a deployment actually starts

  1. 01Choose one high-friction operational question.
  2. 02Identify the minimum required data sources.
  3. 03Connect in read-only mode.
  4. 04Validate the answer against your current process.
  5. 05Deliver one working dashboard, report, or agent workflow.
  6. 06Expand only after the first workflow is trusted.

Walk one workflow with us.

The review is a working session: your question, your systems, and an honest answer about what maps today versus what needs deployment work.

Book a plant workflow review