Main site

Mining AI agent · Process control

Process-Control & Monitoring AI Agent (Processing Plant)

Explain what moved before an excursion, propose a setpoint change with the evidence beside it and check that change against the operating envelope — while the operator and the metallurgist keep every change.

Typical delivery
4–6 weeks
Deployment
Your stack
Setpoint change
Operator only
After launch
Agent Care

What this agent does

Watches the plant, writes nothing to it

In

  1. 01

    Take the excursion, the alarm event and the tag trend as the control system and the historian hold them.

  2. 02

    Read the laboratory results, the unit states, the blend that changed and the work done on the circuit.

Reason

  1. 03

    Assemble what moved before the excursion — which tags, in what order, and which instruments can be trusted.

  2. 04

    Propose a setpoint change against the circuit's control philosophy, with the trends and samples behind it.

  3. 05

    Test the proposal against the operating envelope, the constraints in force and the interlocks holding the unit.

Decide

  1. 06

    Test each reading against its calibration date and its drift and freeze checks, and show what a tag that fails has fed.

  2. 07

    Route anything touching the envelope, a protective layer or the deposition plan to the metallurgist or the superintendent.

Out

  1. 08

    Present the account, the proposal and the envelope check as drafts, with the source tags beside them.

  2. 09

    Draft the shift handover and the unreconciled metallurgical picture, and retain every operator amendment.

Product statement

The agent watches, explains, proposes and checks. Writing a setpoint, changing a loop mode, moving an alarm limit and authorising a deposition stay with the named person.

Example workflow

One excursion, end to end

  1. 1Excursion detectedA tag leaves its normal band, an alarm is raised, or an operator asks what is happening to a circuit
  2. 2Plant data gatheredTag trends either side of it, laboratory results, unit states, the blend that changed and work done on the circuit
  3. 3Account assembledWhat moved, in what order, and which of it the instruments can actually be trusted on
  4. 4Proposal checkedA setpoint change tested against the operating envelope, the constraints in force and the interlocks holding the unit

No human action required

Stages 1 to 4 run without a person in the loop — detection, gathering and the envelope check finish before anyone is asked to read anything. Nothing on the plant changes in that stretch.

5DecisionSplits on analysis confidence and whether the envelope or a protective layer is involved
Inside the envelope, instruments healthy

Reaches the operator as a complete account.

Envelope, interlock or deposition involved

Goes to the metallurgist before it goes further.

Control-room operator or metallurgist

Reads the account, the evidence, the envelope check and what is unresolved, then decides whether anything on the plant changes.

Accept account · Amend · Escalate

Account accepted — handed back

  1. 6Handed to the control roomPresented to the operator as a proposal; no setpoint, loop mode, alarm or deposition is changed by the agent
  2. 7Outcome evaluatedWhat the operator amended, what the shift review concluded, and what the reconciled balance later showed

Amendments

What the operator corrects at the desk is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Writing a setpoint to the control system.
    • Putting a loop in or out of automatic.
    • Starting, stopping or isolating any unit.
    • Acknowledging, shelving or suppressing an alarm.

  • Automation boundary

    Agent acts unaided

    • ✓Explain an excursion and what moved in the hours before it.
    • ✓Propose a setpoint change with the evidence behind it.
    • ✓Test a proposal against the operating envelope and the constraints in force.
    • ✓Draft the shift handover and the unreconciled metallurgical picture.

    Write actions run only inside the approval boundaries agreed during implementation. A setpoint is never one of them.

    • Changing an alarm limit or a trip setting.
    • Overriding an interlock or a permissive.
    • Authorising a deposition, a discharge or a transfer.
  • Signing a metallurgical balance or a production figure.

Example output

One excursion, annotated

Everything the agent proposes is attached to the tags, the samples and the shift it came from.

Excursion account · one circuit, one shift

Illustrative example
Excursion as raised
Rougher circuit out of band
Circuit
Flotation, after a blend change
Envelope rev
Rev 8
Proposed
Setpoint change, for review
Confidence
84%
Setpoint
Not written by the agent

As received

The excursion as the control system raised it, the circuit it sits in and the envelope revision current at that moment.

Evidence used

  • Tag trends either side
  • Shift composite assay
  • Blend and reagent change

Why it stops here

The proposal sits inside the envelope, but the analyser feeding it is past calibration.

What the score decides

Confidence decides how hard the operator reads this, not whether the setpoint moves.

Action

Accept accountAmendEscalate

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value lands

Value 01 – 04

Every excursion the plant raises

From the control system and the historian

03Analysis & proposal

Read the circuit's own limits

Use the control philosophy for that unit, the operating envelope, the constraints in force this shift and the interlocks holding it.

01Approved path

Explain the excursion, not just flag it

The trends, the laboratory results, the blend change and the work done arrive as one account, while the shift still has room to act on it.

02Human review

Send the envelope case up, not out

A proposal that sits outside the envelope, and anything reading against the deposition plan, goes to the metallurgist and the superintendent rather than onto a control-room screen.

04

Build an evidence trail

Retain the tags and samples read, the envelope revision, the proposal, the confidence and every operator amendment — on both paths.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • Control & historian

    • DCS
    • PLC
    • SCADA
    • Process historians
    • OPC UA

  • Plant instrumentation

    • Online analysers
    • particle size
    • Density, flow and level
    • weightometers

  • Laboratory & metallurgy

    • LIMS
    • shift composites
    • Mass balance
    • metallurgical accounting

Agent

Process control & monitoring

  • Explains the excursion
  • Proposes a setpoint
  • Routes to the operator

  • Alarm & shift records

    • Alarm and event logs
    • shift logs
    • Handover records
    • deposition records

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and the control system

Each control wraps the one inside it. A proposal clears every layer before an operator reads it, and writing it to the plant sits outside all six.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeReturn analysis to the control room and the metallurgist if evaluations or production signals degrade.Roll back
L5Change controlRecord the tags and samples read, the envelope revision, the proposal, the named approver and every operator amendment.Record
L4No-write boundaryThe agent holds no write path to a setpoint, a loop mode, an alarm, an interlock or the deposition system, and nothing it raises is annunciated as a plant alarm.Hold
L3Envelope checkA proposal is tested against the envelope revision, the constraints in force and the permissives holding the unit; anything that fails is held and shown held.Stop
L2Instrument healthReadings are tested against calibration date, drift and freeze checks; a tag that fails is marked suspect and the operator is shown what it fed.Test
L1Tag and source checkNumbers carry the tag, the time and the sample they were read from, and what the agent could not resolve is shown as unresolved rather than filled in.Cite
Model coreAccount assembled — what moved, the proposal, the envelope check and confidence
L1 – L2
Decide whether a reading may be trusted
L3
Decides whether a proposal may stand
L4 – L5
Keep the change with a person and the record intact
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the whole account — not only the proposal an operator reads.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the account the operator opens

  1. E1Final-output evaluationWas the account right about what actually moved, and in what order?
  2. E2Step-level evaluationDid it read the current envelope, constraint and control philosophy?
  3. E3Tool evaluationDid it resolve the correct tag, unit and laboratory sample?
  4. E4Envelope recallHow many proposals against the envelope or a protective layer were held?
  5. E5Slice evaluationHow does account quality change across specific excursion cohorts?
  6. E6Business outcomeWhat did the operator amend, and what did the shift review find?

Floor — what the shift review and the balance later showed

Failure modes

Where each failure originates in the agent

Seven failure modes plotted against the five stages of the agent lifecycle.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    2 modes

    • PC-01

      Drifting analyser read as ore

      A slow instrument drift is explained as a change in the feed.

    • PC-02

      Deposition reading read as ordinary

      A pond, beach or freeboard reading outside the plan is shown as normal movement.

    Stage gathers

    Tag trends, laboratory results, states and the envelope

  2. 02 · Analysis

    2 modes

    • PC-03

      Correlation given a direction

      Two tags that moved together are presented as cause and effect.

    • PC-04

      Proposal built on a stale envelope

      The envelope revision used was superseded at the last change.

    Stage proposes

    What moved, in what order, and the setpoint to consider

  3. 03 · Envelope check

    1 mode

    • PC-05

      Permissive not read into the check

      A proposal assumes a unit that a permissive is holding out.

    Stage checks

    The envelope, the constraints and the interlocks in force

  4. 04 · Handoff / write

    1 mode

    • PC-06

      Output adds to the operator's load

      So much is raised beside the alarm system in an upset that it stops being read.

    Stage presents

    The account the operator reads and where it sits

  5. 05 · Change / Version

    1 mode

    • PC-07

      Philosophy change untracked

      An envelope or control-philosophy change reaches the agent with no named approver.

    Stage tracks

    Model, prompt, envelope and philosophy changes

  • Sev-1 · plant or a protective layer is exposed
  • Sev-2 · a wrong account reaches the operator
  • Sev-3 · analysis degrades, more is amended
See the complete Mining failure-mode catalogue

Affected slices

The hours after the ore changes are where this fails

Most of a shift is steady state on a familiar blend and those accounts hold up. The hours after a feed change do not, nor does a circuit whose analyser has drifted, nor a plant in upset. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Hours after a blend change4.8%3.0×Review
Circuits with a drifting analyser3.4%2.1×Review
Plant already in an upset2.9%1.8×Watch
Steady state, familiar blend1.3%0.8×Normal

Bar: amended-or-withdrawn account rate, lift vs. steady-state baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

A setpoint that worked once is not an explanation

Amending the account for one shift leaves the reason it read wrong in place. What changes is the tag it trusted, the envelope, or the rule that raised it.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Operator amendments or shift-review findings move in a cohort.

  2. 02

    Diagnose

    The shift review already named the cause; the question is which tag the agent trusted to get somewhere else.

  3. 03

    Improve

    The tag rule or envelope reference goes through management of change, with an approver.

  4. 04

    Verify

    Replayed over past excursions the plant has already reviewed to a cause.

  5. 05

    Learn

    The excursion is kept for replay and the philosophy note the operators read is amended.

  6. Learn → Detect

    The return edge. An operating envelope or a control philosophy is changed under management of change, never by the agent.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, plant data sources, analysis and envelope logic, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Workflow discovery and automation-boundary definition.Week 1
  2. 02Control-room and shift-handover assessment.Week 1, Week 2
  3. 03Operating envelope and constraint capture.Week 1, Week 2
  4. 04Historian, analyser and LIMS data access.Week 2
  5. 05Excursion detection and account assembly.Week 2, Week 3
  6. 06Setpoint proposals and evidence linking.Week 3
  7. 07Instrument-health and drift marking.Week 3, Week 4
  8. 08Envelope, constraint and deposition checks.Week 4
  9. 09Evaluation suite, envelope recall and regression cases.Week 4, Week 5
  10. 10Operator review and amendment capture.Week 5
  11. 11Handover and metallurgical-picture drafting.Week 5, Week 6
  12. 12Observability, deployment and Agent Care handover.Week 6

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel

Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tierPilotOne circuit, one plantProductionProduction control-room integrationAdvancedMulti-plant / multi-circuit
Introduced at Pilot
Excursion accounts from your own historian✓✓✓
Setpoint proposals with the evidence attached✓✓✓
Operating-envelope and constraint checking✓✓✓
Calibration, drift and freeze tests on each tag✓✓✓
Deposition limits read, never proposed against✓✓✓
No write path to a setpoint, mode or alarm✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Operator review workflow and shift handover—✓✓
Observability and evaluation—✓✓
Introduced at Advanced
Multiple plants, circuits and control systems——✓
Group reporting and enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on the control and historian systems in scope, the number of circuits and plants, analyser and laboratory sources, approval controls and deployment requirements.
Separate from build
Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

  1. 01Your control philosophy and the envelope each circuit runs to

    Operating envelope and constraint capture

    Week 1

  2. 02How an excursion reaches your control room today, and who acts

    Control-room and shift-handover assessment

    Week 1

  3. 03Access to the historian, the online analysers and the LIMS

    Historian, analyser and LIMS data access

    Week 2

  4. 04Your calibration records and which tags you already distrust

    Instrument-health and drift marking

    Week 3

  5. 05Your deposition plan, its limits and who is told when one moves

    Envelope, constraint and deposition checks

    Week 4

  6. 06Excursions you have already reviewed to a cause

    Evaluation suite, envelope recall and regression cases

    Week 4

  7. 07Named operators and a metallurgist to review accounts

    Operator review workflow, then pilot shifts and production validation

    Weeks 5–6

Nothing else is required

Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Phases are drawn over the weeks they actually occupy. Week 5 carries both the evaluation work and the first shifts an operator reads accounts on.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Control philosophy, the envelope and the automation boundary

W2Historian, analyser and laboratory sources wired in

W3Excursion accounts and setpoint proposals on the first circuit

W4Evaluation suite, envelope recall and instrument-drift tests

W5Control-room presentation, supervised shifts and corrections

W6Operators read live accounts beside their own board, then Agent Care starts

Reading the band
The envelope and instrument-health tests come before anything is put in front of an operator. Week 4 exists so that week 5 is not the first time either is checked.
At the end of W6
Operators have read live accounts against their own board and the metallurgist has amended what did not survive the shift review, then Agent Care takes over monitoring.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Mining AI agent

Build a process-control agent around your control room.

Show us one circuit, the envelope it runs to and how an excursion is written up today. That excursion then has two accounts, the shift's and the agent's, and you decide which of them read the instruments and which one guessed.

Typical delivery
4–6 weeks
Deployment
Your control-room stack
Setpoint change
Operator, always
After launch
Agent Care available

Nestack Agents · Process control & monitoring

AGT-MN-07 · Agent Care available after launch

Build a process-control agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your setup

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.