Main site

Mining AI agent · Pit to port

Logistics & Dispatch AI Agent (Pit to Port)

Plan the movement from pit to port — road, rail, stockpile and ship legs — check each load against its permit, mass limits and fatigue rules, and leave every dispatch to the dispatcher.

Typical delivery
4–6 weeks
Deployment
Your stack
Dispatch
Dispatcher
After launch
Agent Care

What this agent does

Plans the movement, does not release it

In

  1. 01

    Take the product ready to move, the stockpile and shed balances, and today's fleet, crews and rosters.

  2. 02

    Read the permits each combination holds, the rail paths given to you and the ship nominations accepted.

Reason

  1. 03

    Build the movement plan leg by leg — road train, load-out, rail path, stockpile and shiploader window.

  2. 04

    Check each load against the combination's mass and dimension limits and the route its permit names.

  3. 05

    Work out what a late train, a closed road or a moved berth does to every leg downstream of it.

Decide

  1. 06

    Flag a plan that only works if a driver runs past a rest rule, and a load outside the cargo specification.

  2. 07

    Route weather closures, permit variations and ship-window changes to the people who decide them.

Out

  1. 08

    Present the plan a dispatcher works from — legs, times, loads, the paperwork each needs and what is missing.

  2. 09

    Retain the sources read, the limits applied, the confidence and every change a dispatcher made.

Product statement

The agent plans, checks and assembles. Dispatch, load release, ship nomination and any change to a route, a mass limit or a rest rule stay with the people who own them.

Example workflow

One cargo, pit to port

  1. 1Demand drawnThe ship nominations accepted, the rail paths held and the product the mine plan says is ready
  2. 2Legs builtRoad-train movements, load-out slots, rail paths, stockpile and shed balances and the shiploader window
  3. 3Loads checkedEach combination against its mass and dimension limits, the route its permit names and the hours the crew has left
  4. 4Knock-on worked outWhat a late train, a closed road or a moved berth does to every leg downstream of it

No human action required

Stages 1 to 4 run without a person in the loop — building, checking and the knock-on work finish before anyone is asked to read anything. Nothing moves in that stretch.

5DecisionSplits on plan confidence and whether a limit, a permit condition or a rest rule is in question
Within limits, permit and hours

Reaches the dispatcher as a complete plan.

Limit, permit or rest rule in question

Goes to the person who owns that decision first.

Dispatcher

Reads the plan leg by leg, the limits applied, the paperwork attached and what is missing, then decides what moves and when.

Accept plan · Re-time · Refer

Plan accepted — handed back

  1. 6Plan handed overHanded to the dispatcher as a draft; nothing is dispatched, released or nominated by the agent
  2. 7Outcome evaluatedWhat the dispatcher changed, what ran late, what was turned back at the port and what needed a variation

Amendments

What the dispatcher re-times or refers is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Dispatching a vehicle or a train.
    • Releasing a load from the load-out.
    • Nominating or accepting a ship.
    • Signing a consignment or dangerous-goods declaration.

  • Automation boundary

    Agent acts unaided

    • ✓Build the movement plan leg by leg to the port.
    • ✓Check each load against the combination's mass limits and its permitted route.
    • ✓Work out what a delay does to the legs downstream.
    • ✓Assemble the consignment paperwork each leg needs, naming what is missing.

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

    • Varying a route, a mass limit or a permit.
    • Setting aside a fatigue or rest rule.
    • Declaring a road or rail corridor open.
  • Committing tonnage or a window to a customer.

Example output

One leg of one movement, annotated

Everything the agent proposes is attached to the leg and the load it came from.

Movement plan · one cargo, one leg

Illustrative example
Leg
Load-out to rail head
Combination
Road train, permitted route
Ship window opens
In 5 days
Proposed
Move to the night path
Confidence
85%
Cleared to dispatch
Not released by the agent

As received

The leg, the combination and the window it has to make — read from the plan and the nomination, not restated.

Evidence used

  • Permit route and conditions
  • Crew hours remaining
  • Rail path held for tonight

What forces the re-time

The daytime run puts the last truck past the driver's rest window.

What the score decides

Confidence decides how hard the dispatcher reads this, not whether the load may move.

Action

Accept planRe-timeRefer

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 load leaving site

From the movement plan and the nomination

03Movement plan

Plan on the limits that apply

Use the permits each combination holds, the paths the rail operator has actually given you, and the window the ship has.

01Approved path

Follow a delay through to the ship

A late train or a closed road is carried through the load-out, the stockpile and the window it puts at risk, while there is still room to move something.

02Human review

Stop a plan that needs a rule broken

A schedule that only works if a driver runs past a rest rule, or a load that sits over a limit, reaches a person before it reaches a driver.

04

Build an evidence trail

Retain the sources read, the limits applied, the plan proposed, the confidence and every dispatcher 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.

  • Fleet & mine systems

    • Wenco
    • MineStar
    • Modular
    • load-out and weighbridge data

  • Transport & permits

    • Access permits
    • route conditions
    • Fatigue records
    • driver rosters

  • Rail & load-out

    • Rail operator paths
    • train order systems
    • Load-out slots
    • train-loading data

Agent

Logistics & dispatch

  • Builds the plan
  • Checks each load
  • Routes what blocks it

  • Port & shipping

    • Terminal line-up
    • ship nominations
    • Consignment notes
    • dangerous goods

  • 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 road

Each control wraps the one inside it. A plan clears every layer before a dispatcher reads it, and the dispatch itself sits outside all six.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeReturn planning to the dispatch desk if evaluations or production signals degrade.Roll back
L5Version monitoringTrack permit, rule-set, timetable, model and prompt changes, and which version produced a plan.Track
L4TraceabilityRecord the sources read, the limits applied, the plan proposed, the named approver and every amendment.Record
L3Dispatcher gateDispatch, load release, ship nomination and any route, limit or rest-rule variation stay with the named owner.Gate
L2Fatigue and hours testPlans are tested against the hours each crew has left; a sequence that only closes past a rest rule fails the test, and the shortfall is named.Flag
L1Limit and permit checkLoads are checked against axle-group and gross mass, dimensions and the permitted route; what fails or cannot be confirmed is held.Check
Model coreMovement plan proposed — the legs, the times, the loads, the paperwork and confidence
L1 – L2
Decide whether a planned leg may stand
L3
Decides who may dispatch and who may sign
L4 – L5
Keep the record and the limits honest
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the whole movement — not only the plan a dispatcher opens.

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

Surface — the movement plan the dispatcher opens

  1. E1Final-output evaluationWas the plan runnable as written, on that route and those hours?
  2. E2Step-level evaluationDid it read the current permit, path and berth window?
  3. E3Tool evaluationDid it resolve the correct combination, path and stockpile?
  4. E4Confidence calibrationDo low-confidence plans actually attract more amendments?
  5. E5Slice evaluationHow does plan quality change across specific movement cohorts?
  6. E6Business outcomeWhat was re-timed, turned back at the port or needed a variation?

Floor — the cargo that actually moves

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

    • LG-01

      Superseded permit read

      A route condition changed since that permit was issued.

    • LG-02

      Stale berth line-up

      The plan is built from a line-up taken days ago.

    Stage gathers

    Permits, paths, berth windows, balances and rosters

  2. 02 · Planning

    2 modes

    • LG-03

      Plan needs a rest rule broken

      The only sequence that works runs a driver past their hours.

    • LG-04

      Delay stops at the load-out

      A late train is not carried through to the ship's window.

    Stage proposes

    Legs, times, loads and the order they must run in

  3. 03 · Limit check

    1 mode

    • LG-05

      Axle limits pass, gross over

      A combination clears each axle group and is over on gross.

    Stage checks

    Mass, dimension, route and hours against the limits

  4. 04 · Handoff / write

    1 mode

    • LG-06

      Stale cargo declaration attached

      A moisture figure is carried over from an earlier cargo.

    Stage presents

    The plan the dispatcher reads and its queue

  5. 05 · Change / Version

    1 mode

    • LG-07

      Timetable change re-times quietly

      A path change moves every leg with no named approver.

    Stage tracks

    Model, prompt, permit and timetable version changes

  • Sev-1 · a legal limit or a duty is put at risk
  • Sev-2 · a wrong plan reaches the dispatcher
  • Sev-3 · planning goes stale, more is re-timed
See the complete Mining failure-mode catalogue

Affected slices

The plans that fail are the ones off the usual circuit

Three things break a movement plan: rain on an unsealed road, a route only just permitted, and a dangerous-goods consignment. The daily rail circuit breaks almost nothing. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Wet-season road movements5.0%3.3×Review
Newly permitted routes3.4%2.3×Review
Dangerous-goods consignments2.7%1.8×Watch
Established rail circuits1.4%0.9×Normal

Bar: re-timed-or-turned-back movement rate, lift vs. rail-circuit baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

A plan that failed is a constraint the agent had wrong

Re-timing the movement leaves the reason it failed untouched. What changes is the constraint it was planned on, the limit checked against, or the source read.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Dispatcher amendments or port rejections rise on a movement cohort.

  2. 02

    Diagnose

    Re-run the plan against what the permit actually said, not what was used.

  3. 03

    Improve

    A named owner approves the corrected constraint before a plan uses it.

  4. 04

    Verify

    Re-planned over past movements whose actual outcome is already known.

  5. 05

    Learn

    The corrected constraint is written into the rules the next plan is built from.

  6. Learn → Detect

    The return edge. A route, a mass limit or a fatigue rule is changed by the people who own it, 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, transport data, planning and limit 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. 02Movement and dispatch process assessment.Week 1, Week 2
  3. 03Permit, route and fatigue rule sets.Week 1, Week 2
  4. 04Fleet, load-out and weighbridge data access.Week 2, Week 3
  5. 05Rail path and port line-up data access.Week 2, Week 3
  6. 06Movement planning and sequencing logic.Week 3
  7. 07Mass, dimension and route checking.Week 3, Week 4
  8. 08Consignment paperwork assembly.Week 4
  9. 09Evaluation suite, limit recall and regression cases.Week 4, Week 5
  10. 10Dispatcher review and amendment capture.Week 5
  11. 11Dispatch-system integration and write-back.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 corridor, one productProductionProduction dispatch integrationAdvancedMulti-site, multi-corridor
Introduced at Pilot
Movement plan built leg by leg✓✓✓
Mass, dimension and route checks per load✓✓✓
Fatigue and hours tested against the plan✓✓✓
Knock-on effect of a delay worked through✓✓✓
Consignment and dangerous-goods paperwork✓✓✓
Dispatch stays with the dispatcher✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Dispatcher review workflow and write-back—✓✓
Observability and evaluation—✓✓
Introduced at Advanced
Multiple sites, fleets and rail operators——✓
Group networks and enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on the number of corridors and products, fleet, rail and port integrations, permit and document complexity, 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. 01How a movement is planned and released on your site today

    Movement and dispatch process assessment

    Week 1

  2. 02The access permits, route conditions and fatigue rules you run under

    Permit, route and fatigue rule sets

    Week 1

  3. 03Access to fleet, load-out and weighbridge data

    Fleet, load-out and weighbridge data access

    Week 2

  4. 04Your rail paths and the port line-up as you receive it

    Rail path and port line-up data access

    Week 2

  5. 05Your consignment and dangerous-goods document set

    Consignment and dangerous-goods paperwork

    Week 4

  6. 06Movements that went wrong, and the ones that nearly did

    Evaluation suite, limit recall and regression cases

    Week 4

  7. 07A named dispatcher to review plans

    Dispatcher review workflow, then pilot plans 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 plans a dispatcher reads.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Dispatch process, permits and the automation boundary

W2Fleet, load-out, rail path and port line-up data wired in

W3Movement planning, sequencing and the limit checks

W4Evaluation suite, limit recall and document assembly

W5Dispatch write-back, supervised plans and targeted corrections

W6A full dispatch cycle run beside the existing board, then Agent Care takes over

Reading the band
Limit checks are drilled in week 4 on movements that have already run. A mass limit read wrong is not a scheduling error, so it is not left for the pilot to find.
At the end of W6
A live dispatch cycle has run beside the existing board and the re-timed plans have been counted against what broke them, 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 dispatch agent around your pit-to-port chain.

Show us the movement you run most, the permits it travels under and who releases a load today. A week later you have that movement planned leg by leg, with the legs no plan can close without a person deciding named as such.

Typical delivery
4–6 weeks
Deployment
Your dispatch stack
Dispatch
Dispatcher, always
After launch
Agent Care available

Nestack Agents · Logistics & dispatch

AGT-MN-05 · Agent Care available after launch

Build a dispatch 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.