Machine Pilot AI
Now in pilot with HMPS and Big Dog

Industrial troubleshooting, grounded in your machines.

Machine Pilot is the factory-floor AI that answers operator questions about your machinery — grounded in your PLC code, plant SOPs and OEM manuals. Every claim is cited. Every safety bypass is refused. Operators get help in seconds, not hours.

Cites every answer Refuses unsafe bypass requests AU data residency
OEE
87%
▲ 6%
Running
24
of 28
Downtime
2h 15m
▼ 18%
Alerts
3
attn
P1
Palletiser 01 — Running
98%
CP
Case Packer 02 — Warning
74%
RC
Robot Cell 04 — Faulted
32%
CITED HMPS 8000 manual · §11.4 · p.34
Trusted by
HMPS — OEM partner
Big Dog Plant — pilot customer
Australian-built · SYD data residency
The problem

Faults cost more than the part — they cost the shift.

When something goes wrong on the line at 2am, your best operator is asleep, the manual is in a drawer somewhere, and the PLC code is a black box. The cost of guess-and-check is measured in lost production, not labour.

01 · Knowledge loss

Senior expertise walks out the door

Every operator who leaves takes years of plant-specific tribal knowledge with them. New starters fumble through faults their colleagues solved in minutes.

02 · After-hours silence

Critical faults happen at 2am

By the time the right person is reachable, the line has been down for an hour. Most plants escalate too late because operators don't want to "wake people up over nothing."

03 · Costly guesswork

Trial and error costs $15K–$80K+ per day

Without a fast, trustworthy second opinion, operators try the same five fixes in different orders. Real fault diagnosis needs the PLC logic, the SOP, and the manual — together.

How it works

Source-grounded answers, in the right order.

Machine Pilot doesn't guess. Every answer is built from your machine's own truth, in a strict hierarchy. The PLC wins over the SOP. The SOP wins over the manual. Contradictions are surfaced, not hidden.

1
PLC code — authoritative for fault-state, interlock, motion

L5X export ingested into a tag graph. When the question is "why is this fault bit set?" the answer comes from the rung that sets it — not a generic guess.

2
SOPs — authoritative for plant-specific procedures

Your cleaning, changeover, and recovery procedures. Versioned and retrievable. Plant vocabulary preserved.

3
OEM manuals & drawings — how the equipment is built to work

Structure-aware chunking of the manuals. Vision-captioned electrical drawings. Indexed by section, page, and component.

Hard safety rule — non-overridable. Machine Pilot refuses any request to bypass, defeat, jumper, or override a safety interlock, guard, e-stop, or safety-rated circuit. No prompt, no tenant config, no user role can change this. Operators get pointed to the right qualified path.
Capabilities

Built for industrial reality, not a chat window.

The features that matter on a plant floor — citations, safety refusals, multi-source reasoning, audit trails — are first-class. Not bolted on.

Cites every claim

Inline citations to section, page, PLC tag, and routine. Operators can verify, supervisors can audit, suppliers can trust.

Safety-first refusals

Hard refusals on guard bypass, interlock defeat, or any safety-rated override — built into the base prompt, not toggleable by configuration.

Multi-source reasoning

One question can pull from PLC, SOP, manual, and drawings in a single answer — and flag where they disagree.

Multi-tenant by design

Each customer's corpus, system prompt, and model behaviour is configured per tenant. One platform, many plants — no data ever crosses.

Audit trail + analytics

Every operator turn, model decision, and source citation is logged. Supplier audit, plant management reporting, regulatory traceability — all out of the box.

Browser, tablet, or chat

Operators interact on the screen they already use. Future: WhatsApp and Slack adapters drop in without changing the engine.

Who benefits

One platform, three audiences.

Operators get answers. Maintenance gets context. Plant managers get visibility. Everyone gets the audit trail.

Operator

Resolve faults in minutes, not hours.

  • Plain-language guidance with HMI button names in bold
  • Photo upload of fault screens, vision-parsed and explained
  • Refuses requests that would compromise safety
  • Time-saved counter visible after every resolution
Maintenance

Walk in with context, not questions.

  • Full transcript + cited sources for every escalation
  • Maintenance schedule pulled from OEM intervals
  • Trending photo-eye / sensor issues flagged before they fail
  • Spare-parts list linked to the relevant page of the manual
Plant manager

OEE, downtime, and resolution metrics, daily.

  • Live dashboard: OEE %, machines running, alerts, downtime
  • Per-shift, per-machine, per-fault-code breakdowns
  • AI-recommended interventions before faults compound
  • Exportable reports for supplier and regulatory audit
~28 min
Average time saved per resolved fault vs. typical escalation path
82%
Faults resolved by AI in the Big Dog pilot, first 30 days
100%
Answers cite their sources — auditable end-to-end
0
Successful safety-bypass attempts — refused 100% of the time
Pilot data from asset 13859 (HMPS 8000 de-palletiser) at Big Dog · stats updated as the pilot extends
Pricing

Pick the tier that matches your plant.

One platform, three tiers, transparent metering. All tiers ship with citation grounding and safety refusals — they're not premium features.

Essential
$890/ month
For single-machine pilots
  • One machine, one site
  • OEM manual + SOP ingestion
  • Browser / tablet operator chat
  • Source citations on every answer
  • Standard safety refusals
  • 500 queries / month
  • Community support
Start a pilot
Enterprise
Custom
For multi-site groups and OEMs
  • Unlimited machines, multi-site
  • All Pro features
  • SSO / SAML + audit log
  • Australian data residency option
  • Dedicated technical specialist
  • Unlimited queries
  • 4-hour critical SLA
Contact sales
Security & compliance

Built with the assumption that your data matters.

Customer / OEM documents are sensitive. We treat them that way from day one — encryption, tenant isolation, hard safety rules, and Australian hosting where you need it.

Bearer-token API auth

Per-tenant API keys with tier scoping. No cookie sessions, no shared credentials, no auth surprises.

Strict tenant isolation

Every database row carries a tenant_id. Every query filters by it. Customer A never sees Customer B's documents, fault logs, or system prompt.

Non-overridable safety rules

The safety refusals are in the base system prompt. Tenant configuration can extend, never weaken. AS 4024 / ISO 13849 aware.

Immutable audit log

Every operator turn, every model decision, every cited source — logged with timestamps. Supplier and regulator-ready.

Australian data residency

Hosted on DigitalOcean Sydney (SYD1). No data leaves the country unless you explicitly extend to a multi-region deployment.

Encryption in transit & at rest

TLS 1.3 for all traffic, AES-256 at rest for conversation logs and ingested documents. Secrets managed in the DO App Platform vault.

FAQ

The questions plant managers and OEMs ask first.

How is Machine Pilot different from ChatGPT or generic AI assistants?

Generic chatbots hallucinate. Machine Pilot is wired into your specific PLC, SOPs and OEM manuals — every answer it gives is grounded in your machine's own documentation, with citations. The model is also configured to refuse any request that would compromise safety (e.g. bypassing an interlock), which generic chatbots do not enforce.

What if our PLC export isn't available yet?

The platform runs in a "manuals and SOPs" mode until the PLC export is available, and clearly tells operators when a question would benefit from PLC context that isn't yet ingested. When the L5X export arrives from your OEM, ingestion is a single job run — no code change.

What does onboarding a new site look like?

For an existing customer adding a second site: drop the new site's documents into their cloud folder, add a tenant row with site-specific configuration, run the ingestion job. Typically under one hour of work. For a brand-new customer: a kick-off call to map their documentation, then 1–2 weeks to first operator demo with their real corpus.

Does the AI ever just guess or make things up?

The system prompt is hardened to require source attribution on every factual claim. If the retrieved chunks don't cover the question, the model is instructed to say so and ask a clarifying question — not to invent fault codes, tag names, or procedures. We monitor citation accuracy as a first-class quality metric.

Can it integrate with our existing tools?

Yes. The API is the product. The browser / tablet wireframe you see in the demo is one interface; Slack, WhatsApp, and embedded-in-your-own-dashboard integrations all hit the same /ask endpoint. The roadmap covers Slack escalations in Phase 2 and WhatsApp field-tech mode in Phase 3.

Who owns the data?

You do. The conversation logs and ingested documents belong to you. We use the platform's logs to improve quality on your tenant only — never to train models that other tenants benefit from. Enterprise tier includes a contractually scoped data-use clause.

Stop fault-finding alone.

We're booking 10-minute discovery calls with plant managers and OEMs through May–June. Walk us through one of your machines, see Machine Pilot answer a real question from your documentation, decide if it's worth a pilot.