The right instruction, the past root cause and the exact spec — found during the shift, with the source attached.
One private knowledge base across your work instructions, machine manuals, maintenance history, quality records, specifications, scanned paper and ERP. Every answer opens the exact passage it came from and tells you whether that document is still current. It runs on a box at your plant, with your data under your control, operated by Olyra as a managed service.
What is Olyra Knowledge Engine for manufacturers? It is a private knowledge base for plants of roughly 50–500 people. It connects to the file server, mailboxes, Microsoft 365 or Google Workspace and your ERP, reads everything including scanned paper in the languages your plants use, and organizes it by categories you define, such as plant, line, machine, product and document type. Operators, engineers, quality staff and AI agents then ask in plain language and get an answer that cites the exact passage, shows whether the source is verified or superseded, and respects each person’s permissions. It runs on a box at your plant or a server in your own cloud account. Your data and keys stay with you; Olyra installs it and operates it as a managed service that keeps it current, secure and supported.
Your plant has written it down. Nobody can find it at 2 a.m.
The procedure exists. So does the 8D from the last time this defect appeared, the supplier’s material spec and the note about why that fixture was changed. They sit in binders, on a file server, in the ERP and in three people’s mailboxes.
“Ask him, he knows that machine.”
The setup trick, the fault that always follows a certain alarm, the reason a parameter was changed. It lives with one long-tenured operator or maintenance lead, and it leaves when that person retires.
“Which revision is the right one?”
A printed work instruction, a PDF on the share and a copy in an e-mail disagree. Someone follows the outdated one because nothing says it has been superseded.
“Haven’t we solved this before?”
A non-conformance opens and the investigation starts from zero. The earlier root cause and corrective action exist, but nobody can find them, so the same analysis is paid for twice.
The same question, again
Torque value, changeover sequence, approved alternate material. It gets asked in WhatsApp or e-mail every week, and the answer depends on who happens to reply.
“It takes months before they can work alone.”
New hires learn by shadowing because the documented process is too hard to navigate. Your most experienced people spend their shift answering instead of producing.
“Where is the evidence?”
An audit date is set and the scramble begins: training records in one place, calibration scans in another, the signed procedure in a binder nobody has opened.
Generic AI tools do not solve this. They cannot read a scanned inspection sheet, they handle local shop-floor language and terminology poorly, they do not see your ERP, and they cannot tell a current instruction from a retired one. And an automation or AI agent working inside a production step needs the exact value now, not a paragraph a minute later.
What a shift looks like once the plant can answer its own questions
Knowledge Engine does not add another place to store documents. It reads the places you already have and makes them answerable.
| Situation | Today | With Knowledge Engine |
|---|---|---|
| An operator needs a setup value | Finds a supervisor, or uses the sheet taped to the machine | Asks in plain language and gets the value with the setup sheet opened at the exact passage, marked verified or superseded |
| A defect reappears | Quality starts a new investigation | A search returns earlier non-conformances, their root causes and corrective actions for the same part, material or supplier |
| A machine throws a fault | Maintenance calls the one person who remembers | The manual section, past maintenance records and related e-mails come back together, each with its source |
| A new hire starts on a line | Shadows a colleague for weeks | Reads the procedure as numbered, cited steps with roles and approval points, and asks follow-up questions against the same sources |
| An auditor asks for evidence | Several people search folders and binders | The procedure, its version history, who can access it and the related records are retrieved from one place |
| An agent needs data mid-task | Waits while a chat model re-reads documents, or guesses | Reads a database prepared in advance, fast enough to use in the middle of a task, and abstains if the evidence is not there |
Chosen for the plant, not the head office
Each capability below answers one of the problems above.
Reads the paper trail
For scanned records nobody can search
- Built-in OCR reads scanned PDFs and images in the languages your documents use
- Inspection sheets, signed procedures and old manuals become searchable text
- Documents, mail, chats, spreadsheets and database records are read together
- Everything is prepared in advance as full text, meaning, a graph of people and things, and structured records
Answers that open the source
For “where does it say that?”
- Every answer cites document, version and exact location
- One click opens the original with the passage highlighted
- Conflicting numbers, dates or revisions between sources are flagged, not hidden
- When the evidence is not there, it says so instead of guessing
Current or superseded, stated
For the outdated work instruction
- Each document is marked verified, stale, superseded or disputed
- An answer that leans on an old document tells you
- Owners are asked to re-check only the content people are using, one click from an e-mail
- Tracks documents, never scores people
Procedures as steps
For SOPs that sit in binders
- SOPs, policies and procedures become versioned, cited steps
- Each step carries its roles, systems, thresholds and approval points
- Findings show where tickets or mail suggest a process is running without its approval step: documents and counts, never people
- Answers approved by your steward take precedence for the questions that matter most
The last time this happened
For investigations that repeat
- Search by meaning and exact keyword together, so a part number and a description both work
- Filter by source, document type, date, person or your own categories
- Staff corrections become precedents the next person or agent can see, once a steward approves them
- View the knowledge base as it stood on any past date
ERP next to the documents
For answers that need the record, too
- Ready-made read-only connectors for TOTVS Protheus, SAP Business One, Senior and Microsoft Dynamics 365 Business Central; connectors for other ERPs such as Oracle NetSuite, SAP S/4HANA, Epicor, Infor and Sage are built and added for each customer
- Customers, orders and invoices appear beside the e-mails and contracts about them
- Spellings of the same supplier or customer are merged into one entity
- Named business figures come from ERP records through definitions your steward manages, with the query shown
People use it through a web console. AI assistants and your own automations connect through the open MCP standard, each under its user’s permissions, and every call is logged. The full capability list is on the Knowledge Engine product page.
Before the person who knows retires
No software reads minds. What Knowledge Engine can do is make sure that everything an experienced person has already written, answered or corrected stays findable after they leave.
- Years of answers are already written. The e-mails where your maintenance lead explained a fix, the notes on a shift report, the comments on a non-conformance: once those sources are connected, they are searchable by meaning and cited like any other document
- Corrections become precedents. When an experienced colleague corrects an answer or a draft, a steward can approve that correction so the next person and the next agent see it
- Approved answers for the questions that recur. Your steward can curate an answer to the question that is asked every week, and that answer takes precedence
- Gaps become a to-do list. Questions the knowledge base could not answer are reported, so you know which know-how still needs to be written down while the person is there to write it
- Who owns this? When the evidence is missing, the engine can point to the owning role or your data steward, so the question reaches a person instead of a guess
This reduces key-person risk and supports a cleaner handover. It works on what has been captured in a connected source; know-how that exists only in someone’s head still has to be written or recorded once.
Exact data in the middle of a task
A person can wait a minute for an answer. An agent preparing a quote, checking a purchase order against a specification or drafting a reply to a customer complaint cannot.
- Prepared in advance. Your data is built into a high-speed database in several formats, so a lookup reads what is already there instead of re-reading documents through a chat model
- Fast enough to use during the shift. Lookups can return in as little as a couple of milliseconds. Results vary with hardware, data volume and query type
- Your process, cited. An agent reads your procedure as versioned steps with roles, thresholds and approval points instead of improvising from general knowledge
- Knows when to stop. It reports insufficient evidence, flags conflicting revisions and routes the question to the owning role
- Checked before it is sent. A grounding check reviews a draft sentence by sentence against your records and shows a wrong number, date or revision with both values. It never edits or blocks on its own
- Only what its user may see. Permissions are enforced in the database, and every agent call goes to the audit trail
Agents and automations that speak the open MCP standard connect under a user’s identity or a service account you approve. Olyra’s pre-built AI agents use the same knowledge layer.
Evidence you can retrieve, not reconstruct
Quality-system audits ask the same things each time: show the current procedure, show who can change it, show the record. Knowledge Engine keeps those answers retrievable.
Current procedure, with history
Each document carries a verification status, and the knowledge base can be viewed as it stood on a past date, so you can show what the instruction said at the time.
Who can see what, and why
Access reviews show which people can see a document and the reason. Roles say what a person may do; visibility by your own categories, such as plant or department, says what they may see.
A record of use
Sign-ins, searches, document views, assistant calls, label changes and exports go to a tamper-evident audit trail. Sensitive settings need two people to approve.
Past findings, findable
Non-conformances, corrective actions, 8D reports and audit findings that are stored in connected sources come back by part, supplier, material or description.
Retention under your rules
Keep-for periods by label or source. The system proposes what is due, and nothing is erased without approval.
Supplier and customer requirements
Material specifications, customer drawings and requirement documents sit next to the ERP records for the same supplier or customer.
Knowledge Engine supports the evidence side of a quality system. It does not certify your organization, and it does not guarantee the outcome of any audit. Nothing on this page is legal, tax or financial advice.
For plants that run Lean, Six Sigma, EOS or an ISO-style system
These disciplines share one assumption: the process is documented, people can find it, and what they find is current. That assumption is where many programs quietly fail.
| Discipline | What it depends on | How Knowledge Engine supports it |
|---|---|---|
| Lean / standard work | The current standard is the one people follow | Work instructions become cited steps; superseded versions are marked, so the standard in use is the standard on file |
| Six Sigma / problem solving | Past analyses and data are available to the next project | Earlier root-cause work, corrective actions and related records are searchable by meaning, with sources |
| EOS | Core processes are documented and followed by all | Documented processes are readable as steps with roles, and findings show where practice appears to skip an approval point |
| ISO 9001-style quality systems | Controlled documents and retrievable records | Version history, verification status, access reviews and an audit trail for the documents and records that have been ingested |
Knowledge Engine does not implement or certify any of these methods. It makes the documentation they rely on findable and current. If you also want one trusted set of plant numbers, that is our Operational Intelligence work, and the two fit together: figures on one side, the knowledge behind them on the other.
What you would otherwise try
Each of these is a reasonable choice for some plants. Here is how they typically compare for a manufacturer whose knowledge is spread across a file server, an ERP, mailboxes and paper.
| The alternative | Where it typically shines | Where it typically falls short on the floor | What Knowledge Engine does |
|---|---|---|---|
| Tidy up the file server or SharePoint | No new system; good discipline in any case | Keyword search inside one system; ERP, mail and scans stay separate, and nothing says which revision is current | One cited answer across sources, with the current version marked |
| The document module of your QMS or ERP | Controlled documents with formal approval workflows | Covers the documents that were put into it; mail, maintenance notes, manuals and older scans usually stay outside | Reads those sources alongside your controlled documents. It complements the QMS and does not replace it |
| Copilot bundled with your office suite | Already in the tools office staff use | Centered on that suite’s content; follows today’s permissions, oversharing included | Adds ERP records, file servers and scans, checks permission hygiene before go-live, and can serve Copilot-style tools through MCP |
| ChatGPT or Claude over uploaded files | Fast to start on a handful of documents | Little or no link to your permissions, ERP or audit trail; answers rarely say whether a source is out of date | Permission-checked, cited answers with freshness on every source and a record of who asked what |
| Build your own | A capable team can show a demo quickly | It sees only the sources it was wired to, so it cannot tell a current, approved instruction from a superseded one, and it answers confidently either way. A demo also often re-reads documents for every question, which can take a minute or more. Then come OCR, permissions, sign-in, audit, backups and someone accountable when it breaks | Connects all your sources and records how current and how trusted each one is, so an answer comes with evidence of whether it is right. Data prepared in advance, those parts included, and Olyra installs, updates and operates it |
| A wiki or knowledge-base tool | Clean pages when someone writes them | Depends on people rewriting what already exists elsewhere, and goes stale when they stop | Reads the originals where they live, so nothing has to be rewritten first |
Your data stays at your plant, under your control. Olyra keeps it running.
Drawings, formulations, customer requirements and supplier terms are exactly what a manufacturer does not want in a shared cloud service.
| Yours | Olyra’s ongoing service |
|---|---|
| The data, the knowledge base and every export | Setup: installs and hardens the system, connects your sources, and builds the categories and residency rules with your data steward |
| The recovery keys, handed over at go-live | Runs updates in your maintenance window, with automatic rollback if a health check fails |
| The box at your plant, or the server in your own cloud account, where it runs | Monitors and security-patches the system, and supports it under an agreement with published response times |
| The accounts and addresses where AI models run | Adds connectors and improvements, and delivers evaluation reports on your own questions |
| The decision to approve each support session | Holds no copy of your data, and works on the box only in sessions you approve |
The box and your data are yours. What Olyra provides is the ongoing service that keeps the knowledge accurate, current, secure and growing. Every AI model is reached at an address you control: your own GPU machine, a cloud service under your own account, or the box itself. Olyra does not host AI models for you. Residency rules decide what may leave the box at all, by where content came from or by what it is, and if in doubt the engine blocks and logs the decision. More detail is in our Trust Centre.
One plant today. A second line, a second site, a plant that cannot stop.
Machines are added as needs grow. Nothing is reinstalled or reloaded.
| Stage | What you add | What it gives a plant |
|---|---|---|
| Single box | Nothing | Everything on this page on one machine, with encrypted, tested backups |
| Failover pair | A second box | If the first box fails, the second takes over automatically, for plants where a shift cannot wait |
| Cluster | Three or more boxes at one site | Shares heavy load and keeps serving if a member is lost |
| Several plants | A replica at each plant | One master site; each plant serves its own people locally |
| Disaster-recovery standby | A box at a second site | A ready copy far from the first, brought up if a whole site is lost |
Several plants, one source of truth. One site is the master. When a plant has local documents to add, its replica sends them to the master, the master reads and organizes them, and the plant receives them back at the next replication. A plant therefore has its own new data after replication, not before, and every site stays consistent. How much each stage adds depends on hardware, data volume and configuration, and we size it with you.
Cited answers to your own questions on day 3
We start with the one source that holds the most value, often the quality share or the engineering file server, never everything at once.
Install and connect one source
The box goes in, sign-in is federated with your identity provider, and the single richest source is connected.
Your questions, answered
Ten of your real plant questions are asked and answered in front of you, with citations that open the source.
More sources, first assistant
A starter set of categories is in place, more sources are connected, and a first assistant does one narrow job end to end.
First evaluation report
Measured results on a golden set of your own questions, your first business metrics and a trained data steward.
First ingestion is real work: your documents are read, extracted, classified and built into a high-speed database in several formats, which takes time and AI processing. That is why there is no free trial on your own data. You can see a demonstration box running on sample data before you commit. Remaining sources, classification review, residency rules and roll-out to users follow over the next four to six weeks. If you are not yet sure where knowledge sits among your priorities, an AI Operations Audit is the place to begin.
Who it is for, and who it is not for
- Manufacturers and operations-heavy producers with roughly 50–500 people, in Brazil or the US
- Plants where the exact detail matters: tolerances, sequences, materials, revisions, approvals
- Teams whose records are spread across a file server, an ERP, mailboxes and scanned paper, in one language or several
- Companies with experienced people close to retirement and new hires who need to become productive sooner
- Plants putting AI agents or automation into daily operations that need those agents to be right and to show their work
What it is not. Knowledge Engine does not control machines, and it does not replace your MES, ERP or QMS. It reads from them, read-only, and from the documents around them. It checks only what has been ingested, so a document that was never connected cannot be cited. And if your need is a chat assistant over a few hundred files, a lighter tool will serve you well, and we will say so.
Questions manufacturers ask us
Our documentation is a mess. Do we have to clean it up first?
No. The engine reads the sources as they are, proposes starter categories from your own documents and industry templates, and holds anything unclear as Unknown in a review queue for your steward. Duplicate and conflicting versions are part of what it surfaces: conflicting numbers, dates and revisions are flagged, and documents can be marked superseded. Cleaning up becomes a guided task on the content people use, not a project that has to finish before anything works.
Can it read our scanned inspection sheets and old manuals?
Scanned PDFs and images are read with built-in OCR in the languages your documents use and become searchable text with citations back to the page. Results depend on scan quality: a faint or handwritten sheet will read less well than a clean typed one, and we check this on your own samples early in the engagement.
Does it work with TOTVS Protheus, SAP Business One, Senior or Microsoft Dynamics 365 Business Central?
Yes, through ready-made read-only connectors. Connectors for other ERPs, such as Oracle NetSuite, SAP S/4HANA, Epicor, Infor and Sage, are built and added for each customer. ERP records such as customers, suppliers, products, orders and open items appear next to the documents and e-mails about them. The engine never writes to your ERP.
Will it replace our QMS, MES or ERP?
No. Those systems remain the system of record and the place where work is controlled and approved. Knowledge Engine reads from them and from everything around them, so people can find and trust what is already there. It does not control machines or production.
Can operators use it on the floor, in their own language?
Yes. People ask in plain language through a web console, in English, French, Spanish, German, Portuguese and more, and each person sees only what they are permitted to see. Answers open the original document at the cited passage, so an operator can check the source in one click.
What about the know-how that is only in people’s heads?
The engine works on what has been captured in a connected source. Much more is captured than most plants expect: years of e-mail answers, shift notes, comments on quality records. For the rest, unanswered questions are reported as knowledge gaps, which tells you what to write down while the person is still there, and approved corrections become precedents.
Will our drawings and formulations leave the plant?
Only if a rule you set allows it. The knowledge base lives on a box at your plant or a server in your own cloud account, on encrypted storage, and Olyra holds no copy. Residency rules decide what may go to a cloud AI service or be handed back to an outside assistant. With the GPU configuration, AI models run on the box and restricted content is processed locally.
Will it help us pass our ISO audit?
It makes evidence easier to retrieve: current procedures with version history, access reviews, retention rules and a tamper-evident audit trail. It does not certify your organization and cannot guarantee an audit outcome, which depends on your quality system and your auditor. This is not legal advice.
Can we try it on our own data for free?
No. Bringing your data in means reading, extracting and classifying it and building a high-speed database in several formats, which takes real time and AI processing. We can show you a demonstration box on sample data, and the engagement starts with one high-value source so you see cited answers to your own questions on day 3.
Bring ten questions your plant cannot answer quickly today
In a 30-minute briefing we look at where your knowledge lives, which source would show value first, and what the first 30 days would involve at your plant. You leave knowing how to get it running.
Book a plant knowledge briefing
Prefer to read first? The full product description is on the Knowledge Engine page, and our wider work with plants is on AI for manufacturing.
Find out what AI can actually do for your operations.
Start with an AI Operations Audit — a fixed-scope diagnostic that maps your workflows, scores your AI readiness, and hands you a prioritized 120-day roadmap. If we don't find real opportunities, you'll know that too.
Request an AI Operations Audit → Prefer to talk first? Book a 30-minute scoping call.