GUS.ai
Answers for your cheese plant.
Ask in plain words, the way you'd ask a colleague. GUS.ai answers from your SOPs and manuals, your PLC program and your live vat data. It shows you its sources so you can check them, and it never writes to your controls.
What is GUS.ai
Three knowledge sources.
One conversation.
A plant's know-how lives in three places: the document library, the live process data and the PLC program. Getting an answer usually means knowing which one to check, or who to ask. GUS.ai searches all three in one conversation and shows you its sources.
Your documents
SOPs, P&IDs, equipment manuals, recipe specs, parts lists, reference handbooks and operator screens. Every passage GUS.ai uses keeps its source and location, so you can check the original.
Live process data
Current values, trends, alarms and batch state from your vat's instruments, collected through APT's ADX data system. Read-only.
Your PLC program
Allen-Bradley ControlLogix programs exported as L5X (controller tags, UDTs and recipe references) plus the FactoryTalk alarm export, searchable in plain English. GUS.ai reads the exported files, not the running controller.
Chapter 01
What
GUS.ai does.
Everything GUS.ai can do today.
Capabilities
What GUS.ai can do.
Built for the questions plant engineers and operators actually ask.
Document answers
Search SOPs, manuals, P&IDs, recipe specs and reference handbooks in plain words. Answers cite their source, and PDF citations open the original at the cited page.
About 5,300 indexed passages · 15 document pipelines
Drawings
GUS.ai reads P&IDs and figures, not just text, and can look closely at the region of a drawing around a tag before it answers.
P&IDs · figures
Live data and batch history
Current values, trends and alarms for the instruments mapped for your process. Step timing, batch-to-batch comparison and spec-vs-actual deviations. Batch boundaries are worked out from the vat's own recipe and step tags, so no MES is needed.
Live tags · trends · batch history
PLC program
Ask about the tags, UDTs and alarm definitions in your exported L5X program and FactoryTalk alarm export, in plain English.
Allen-Bradley ControlLogix · L5X
Process health
Flags unusual readings, finds tags that move together, and traces what a failed sensor, valve or pump would affect, using a model of how the vat's equipment depends on each other.
Anomalies · correlation · impact and weak-point checks
Web and email
Ask from the web app or by email. Email answers use a smaller, read-only set of tools. Optional email notices when a batch or CIP cycle starts or ends.
Web · Email · batch and CIP notices
Operator screens
HMI operator screens go into the knowledge base too, so you can ask what a screen shows and where a value comes from.
FactoryTalk View SE · Optix exports supported
Audit trail and sign-in
Answers are written to a hash-chained audit log, with daily digests and CSV export. If a write ever fails, the answer still goes out, the miss raises an alert, and nothing is backfilled. Microsoft Entra ID sign-in is supported.
Hash-chained log · CSV export · Entra ID supported
Visualization
See your plant,
your program, your data.
Six purpose-built views. On most of them you can click an item and ask GUS.ai about it.
Knowledge map
How your SOPs, manuals, drawings and references relate, built automatically from the documents.
Equipment dependency map
A model of how the vat's sensors, valves, pumps and process steps depend on each other. It is the basis for the impact and weak-point checks.
Data flow
How data moves from instruments and the PLC through APT's ADX data collection to GUS.ai.
Live data explorer
Watch vat values stream in real time while the live feed is running. Read-only.
Live vat view
A real-time overlay of live process tags. Watch a batch run alongside GUS.ai's answers about it.
Causal reasoning panel
19 ready-made questions on impact, likely contributors, weak points and process health, plus your own.
Chapter 02
Built on
serious AI.
Right model for the right job, with safeguards on every channel.
AI architecture
Right model for
the right job.
Claude for reasoning, OpenAI for speech recognition, plus dedicated retrieval models for document search, each chosen for a specific task, balancing reasoning depth, latency, and cost.
Powered by Claude
Claude Opus
Highest reasoningCore agent brain
Runs the bounded tool-use loop in deep mode, the web default, with extended thinking — deciding which documents to search, which tags to query, and how to synthesize evidence into answers. Also powers the interactive code, knowledge, docs, and dependency graph conversations.
Claude Sonnet
Low latencyFast mode
Runs the same tool-use loop in fast mode with minimal thinking, where latency matters most. Also answers text and phone queries where those channels are enabled.
Claude Haiku
Low latencyReal-time descriptions
Generates instant purpose descriptions for graph nodes on click. Optimized for sub-second latency with server-side response caching.
Mid-query OpenAI failover keeps GUS.ai answering through a Claude-side outage
Powered by OpenAI
Whisper
Speech recognitionSpeech-to-text
Transcribes spoken questions into text for the agent pipeline. Operators can ask GUS.ai hands-free from the plant floor from the kiosk or browser microphone, and hear spoken answers through a swappable text-to-speech provider.
Security & trust
Built for the
plant floor, not a sandbox.
An AI co-pilot for industrial work has to earn the trust of a control engineer before it earns the trust of a CEO. These are the safeguards we ship by default.
Evidence-mandatory architecture
Answers cite their evidence, and thin support is flagged. Tool calls are logged to a tamper-evident chain. Every record in it is independently verifiable.
Read-only. No write path to your controls.
GUS.ai never writes to your PLC. The agent has no tool that can modify a setpoint, force a tag, or change a recipe. Plant safety is enforced at the tool layer, not by prompt.
Tamper-evident audit trail
Queries, tool calls, and responses are appended to a SHA-256 hash-chain — break a record and the whole chain fails to verify. Coverage is fail-open on availability: if the audit store is unreachable you still get your answer, and the missing record raises a security event — or, in a total audit-database outage, an error-monitoring alert. Daily digest emails. CSV export.
Boot-time integrity manifest
Config, tool schemas, safety regexes, and the sandbox boundary are SHA-256 manifested at startup. Every boot logs the hashes and exposes them to admins, so drift between two deploys is visible on comparison.
Emergency kill-switch
Admin-gated endpoint halts the whole co-pilot: web queries return 503, and email (and text and phone, where enabled) get a refusal instead of an answer.
Per-channel tool allowlist
Each channel gets only the tools it needs; text and phone, where enabled, get a smaller read-only subset. Email is scrubbed for prompt-injection attempts at the gateway. Defense in depth.
Encrypted in transit
TLS on every public endpoint (TLS 1.3 measured on the app and API). This site also negotiates hybrid post-quantum key exchange; the app and API do not yet.
Chapter 03
Where it
goes from here.
Today's deployment is the beginning, not the destination.
Roadmap
What's coming next.
GUS.ai is actively evolving. Each milestone is informed by real plant-floor feedback — not a market plan.
Built, not yet switched on
Supervisor-approved agents
Turn a recurring job, such as a shift handoff or an end-of-make report, into an agent. A supervisor approves the agent before it can run, and approves each email before it goes out.
Built, in testing
Plant memory
Remembers plant-specific context between conversations. Built with its own safety test suite, and switched off until it is cleared for use.
Research
Early warning
Catching equipment wear and heat-exchanger fouling before they raise an alarm. It will be tested on historical batch data first.
Planned
More PLC platforms
Today: Allen-Bradley ControlLogix. Next: Siemens, Beckhoff and Schneider programs, answered the same way.
Planned
Beyond cheese
The same approach for other food processes.
Built, not yet switched on
Phone and text access
Ask GUS.ai by phone call or text message, with a smaller read-only tool set.
Planned
Mobile plant-floor app
A phone and tablet app built for the plant floor.
Planned
Graph-aware search
Today GUS.ai can follow equipment connections to the matching manual passages. Next: following the links between documents themselves.
About APT
Process engineers
who build AI.
Advanced Process Technologies has spent 26 years engineering dairy and food-processing systems for some of the largest plants in North America. GUS.ai is what happens when that domain knowledge meets agentic AI.
We don’t hire AI generalists and hope they figure out cheesemaking. We started with a master cheesemaker and wired AI around what he already knows.
GUS.ai is named for Mark Gustafson, our master cheesemaker. It doesn’t replace people like Mark. It helps every shift find what the plant already knows, in its SOPs, its manuals, its program and its data.
Quick facts
- Founded
- 2000
- Ownership
- Employee-owned (ESOP)
- HQ
- Cokato, Minnesota
- Patents
- 3 U.S. patents on the ACV
- Reference deployment
- Advanced Cheese Vat
- Named for
- Mark Gustafson, master cheesemaker
See it on a real vat
Bring GUS.ai to your plant.
A 30-minute walkthrough of the live Advanced Cheese Vat deployment. We’ll show you the agent in action against real plant data, not a sandbox demo.
Greg McMillan · [email protected] · APT, Cokato MN