Skip to content

Docs & Knowledge Center

Ask your manuals in plain language and get answers with sources. One library for manuals, SOPs and work instructions, attached to the assets they cover and used by every AI feature in OpexMX.

Updated

Watch it in the product

Keep manuals, SOPs and work instructions in one OpexMX library: folders, written documents, assignment to all machines, an equipment type or one machine, and the Documents tab on every asset.

Find it: Knowledge Center in the sidebar. In the OpexMX Field app: Me → Manuals & SOPs.

Sound familiar?

  • "Where's the manual for this machine?"
  • "How did we fix this last time?"
  • "The guy who knew this retired last month."

The Knowledge Center keeps manuals, procedures and know-how in one library, attached to the equipment they describe. Technicians can open a document from the asset, checklist step or training lesson that needs it. Or they can skip the document and ask: the AI reads your own manuals and answers with the document it came from.

OpexMX Knowledge Center library table listing manuals and guides with type, size, and created dates beside a folder tree.


Ask Your Manuals

Open AI Chat in the sidebar, or the copilot in the OpexMX Field app, and ask the way you would ask a senior technician:

  • "What's the torque spec for the coupling bolts on the transfer pump?"
  • "Compressor 2 trips on high discharge temperature. What should I check first?"
  • "What's the LOTO sequence for the main conveyor?"

Demo data: a fictional compressor manual. Waiting time is sped up where marked.

The answer comes from your own manuals, SOPs and written documents, not from the internet. Each answer lists the documents it used as source badges, and you can click one to open the document. When nothing in the library covers the question, the assistant says so instead of inventing a procedure.

AI Chat answer to "Air Compressor #1 trips on high discharge temperature. What should I check first?", listing the checks in the order the compressor manual gives them, with source badges for the manual and a knowledge article.

  • Asks in your language: it replies in the language you wrote in, so a question in Indonesian gets an answer in Indonesian, with parameter names and asset codes left as they are.
  • Focus on one manual: type @ and pick a manual to keep the answer to that document.
  • Knows the asset: when you name an asset, it searches that asset's own documents, its equipment type's manuals and plant-wide procedures together.
  • Combines documents with past fixes: for troubleshooting questions it also searches the problem library for known faults and their repair steps, so you get the manual's spec and the fix that worked last time.

How an answer is found

This is retrieval-augmented generation (RAG): the AI looks up the passages that answer the question first, and only then writes a reply from them.

  1. Read. PDF text is extracted on upload by pdftotext-rs, our own extractor written in Rust and released as open source (Apache-2.0). It keeps the reading order of multi-column manuals and handles Chinese, Japanese and Korean fonts. It runs as WebAssembly in the cloud edition, with large PDFs going to a container, and as a native binary on-premise.
  2. Split into passages. The text is cut at paragraph breaks into passages of a few hundred words, with a short overlap so a step that crosses a break isn't lost. A new passage starts at each heading that begins with "Chapter" or "Section". Each passage is tagged as a procedure, specification, PM schedule, troubleshooting or table content.
  3. Index. Every passage is turned into an embedding for semantic search, so "pump won't start" finds a passage titled "Motor fails to energize". Written documents are indexed the moment they are saved; images are indexed by their title and description.
  4. Search. Your question is searched in several forms, for example with filler words and negations removed, because "why doesn't the pump start" is phrased very differently from the manual text that answers it. Results from all forms are merged.
  5. Re-rank. A re-ranking model scores each candidate passage against the question and puts the most relevant first. Weak matches are dropped.
  6. Answer with sources. The best passages go to the language model, which writes the answer and cites them.

Want to see what OpexMX will read from a manual before you upload it? Drop it into the free Manual Text Checker; the file never leaves your browser.

One library behind every AI feature

The same index feeds the other AI features in OpexMX, so every manual you add makes all of them better:

FeatureWhat it takes from your documents
AI Chat (web) and the Field app copilotAnswers with cited sources
Suggest with AI in Problem ManagementA draft problem, root cause and repair steps based on the asset's manuals
Root cause analysis in Reliability EngineeringManual limits and procedures as evidence, alongside failure history and sensor data
Manual-to-PM extraction (below)Suggested PM schedules and threshold rules
MCP serverThe search_knowledge tool for your own AI agents

Where it runs

In the cloud edition, indexing runs on Cloudflare: Workers AI for the embeddings and re-ranking, and Vectorize for search. On-premise, passages are stored in PostgreSQL with pgvector, and embeddings and answers come from the AI endpoint you configure. That endpoint can be a model running inside your own network. Re-ranking is available in the cloud edition only. On-premise, results keep their semantic-search order. Admins choose the AI provider and models under Settings → AI Config.

Your documents are only searched for your own organization. A document that fails to index is flagged in the library and can be re-indexed with one click.

What Goes In

TypeHow you add itLimit
PDFUpload: OEM manuals, SOPs, drawings, datasheets50 MB
ImageUpload: JPEG, PNG, GIF or WebP photos and diagrams10 MB
Written documentWrite Document: a rich-text editor with headings, lists, tables and images—

Written documents suit short work instructions and lessons learned. Photos inserted into one can be annotated with arrows, boxes and text, so "loosen this bolt" points at the bolt.

Organize It

Folders

Build a folder tree that matches how your team thinks: by area, by equipment family, by document kind. Folders can be nested, renamed, reordered and moved, and documents can be moved between them.

Assignment: who a document applies to

Every document has one of three scopes:

  • All assets: plant-wide material, like LOTO procedures or a PPE matrix
  • Equipment type: every asset of that type, like the manual for all your screw compressors of one model
  • One asset: a specific machine's modifications, wiring or history

The asset's Documents tab then shows everything that applies to it, from all three levels, so nobody has to know which folder a manual was filed under.

Where Technicians See It

  • Asset detail: the Documents tab lists every document for that asset
  • Checklists: a checklist step can link a document, shown next to the step while the checklist runs
  • Problem library: the corrective sub-actions for a known problem can link the procedure to follow, shown during problem identification on a ticket
  • Training: course lessons can open a linked document
  • OpexMX Field app: browse by folder, search by title, and open PDFs, images and written documents full-screen

From Manual to Maintenance Plan

Documents assigned to an equipment type or an asset are read for maintenance intervals and operating limits once they are indexed:

  • PM schedules: "replace oil filter every 2,000 hours" becomes a suggested preventive schedule for the asset. Manuals with suggestions waiting appear in the Needs Review card at the top of the Knowledge Center.
  • Threshold rules: "discharge temperature must stay below 110 °C" becomes a suggested rule that can raise a ticket when a reading crosses it.

Manual workspace for a compressor manual, PM Schedules tab: six maintenance schedules extracted by AI, the first a daily oil and pressure check with its tasks and a proposed checklist, and a Create 6 Schedules button.

Nothing is created until a person reviews and confirms each suggestion. Open the manual to see its extracted passages, the suggested schedules and rules, and the PDF side by side. Plant-wide documents (scope All assets) are searchable but not mined for schedules.

Who Can Do What

PermissionAllows
knowledge.readBrowse, open and search documents (web and Field app)
knowledge.createUpload files, write documents, create folders
knowledge.updateEdit, rename and move documents and folders
knowledge.deleteDelete documents and folders

By default, Admins and Engineers have all four; Supervisors, Technicians, Operators and Warehouse staff can read. Adjust it per role in User Management.

Getting Started

  1. Create a few top-level folders: safety, equipment manuals, SOPs, lessons learned
  2. Upload the manuals for your critical assets and assign each to its equipment type or asset
  3. Review the Needs Review card and confirm the PM schedules and rules that match how you run the equipment
  4. Link procedures to checklist steps in your most-used checklists
  5. Ask AI Chat a question you already know the answer to, and check that it cites the right document

Tips

  • Scanned PDFs without a text layer are not OCR'd in the Knowledge Center, so their pages won't be searchable. Run a manual through the Manual Text Checker first; if it comes back empty, get a text-based PDF from the manufacturer or run OCR on it before uploading
  • Give documents descriptive titles: the library search matches titles, and image documents are indexed by title and description only
  • Assign at the equipment-type level when a manual covers many identical machines; use the asset level only for what is unique to one unit
  • Split a long SOP into short written documents, one per task, so a checklist step can link exactly the part a technician needs
  • Write a short lessons-learned document after an unusual repair: AI Chat can use it the moment you save it
  • Ask specific questions. "Torque for the coupling bolts on P-101" finds a better passage than "pump bolts". Naming the asset adds its own manuals to the search

Related