What it is
A search box that knows your manuals.
You turn equipment manuals into custom AI search by converting your manuals, service bulletins and wiring diagrams into searchable text, loading them into a private search system and letting techs ask questions in plain language from a phone or tablet. The system finds the right pages and the AI writes a short answer from them, with a link to the source page.
Think about how your techs find information today. They scroll a 120-page PDF on a phone screen. They dig through a binder in the truck that may or may not have the right revision. They search the manufacturer’s site from a rooftop. Or they call the office, or the one senior tech who knows that unit, and wait for a callback while the customer watches.
A custom search system cuts that to a question and an answer. The tech types or says what they need, reads the answer and taps through to the page in the manual to confirm. It’s a better way to look things up, built on the documents you already own.
What techs ask
The questions it answers best.
Manual search works best on questions that have one right answer sitting somewhere in a document. These are illustrative examples of the kind of thing techs ask:
- What does a flashing error code mean on this model?
- What is the gas valve outlet pressure spec?
- What is the torque spec for this fitting?
- Which replacement part number fits this serial range?
- What is the startup sequence after a board swap?
- What does the latest service bulletin say about this issue?
- Where is the reset switch on this unit?
- What does our own install checklist require here?
It is weaker on questions that need judgment on site: why a system keeps short-cycling when the specs all check out, or whether a repair is worth doing on a twenty-year-old unit. The manual can inform those calls, but your tech still makes them.
How it works
Search first, then answer.
The method behind manual search is called retrieval-augmented generation, or RAG. Instead of relying on what an AI model learned in general, the system searches your documents for the pages that match the question and hands those pages to the model, which writes its answer from them. You don’t retrain the model on your manuals, so adding a new manual is as easy as adding a file. Read AWS’s RAG explanation
- 1
Collect
Gather the manuals, bulletins and diagrams for the equipment you service most.
- 2
Convert
Turn PDFs and scans into clean text, with text recognition for scanned pages.
- 3
Index
Split documents into sections and tag each with brand, model and revision.
- 4
Ask
A tech asks a question in plain words from a phone or tablet.
- 5
Find
The system pulls the sections that best match the question and the model.
- 6
Answer
The AI writes a short answer from those sections and links the source page.
What to load
Start with the documents techs reach for most.
Don’t upload every file you own on day one. Ask your techs which equipment they service most and which documents they hunt for, and start there. A smaller, clean collection gives better answers than a big, messy one.
Installation and service manuals
The core collection, for the brands and models you install and repair most.
Service bulletins
Manufacturer updates that change a procedure or flag a known fault.
Wiring diagrams and parts lists
Linked as images where text conversion would lose the detail.
Your own procedures
Install checklists, startup steps and company rules your techs follow.
Warranty terms
What is covered, for how long, and what voids it.
Fixes your team has found
Notes from senior techs on problems the manual doesn’t cover well.
Keep revisions straight. Manufacturers update manuals, and two versions of the same document can disagree. Tag each file with its revision, retire the old version when a new one arrives, and decide who on your team owns the collection.
The hard parts
Where manual search gets tricky.
Scanned and faded pages
Text recognition handles clean scans well. Faded copies, handwriting and stamps need a person to check the converted text.
Tables and diagrams
Spec tables can scramble during conversion, and diagrams don’t turn into text at all. Both need special handling, often linking the original page image.
Similar models
Two furnaces from one brand can share most of a manual and differ on one critical spec. Tagging by model and serial range keeps answers on the right unit.
Spotty signal
Basements, crawlspaces and rural sites lose connection. Plan for it before rollout, not after the first complaint.
Keep it accurate
Every answer shows its source.
The rule that makes manual search safe is simple: every answer links the manual and page it came from. If the system can’t find an answer in your documents, it says so and doesn’t guess. Techs learn to read the answer, tap the source and confirm, the same way they’d check a spec with a coworker.
Be strict on safety-critical work. Gas pressures, electrical clearances, refrigerant handling and anything covered by code should always be confirmed against the manual page itself, not only the summary. The search tool finds the page faster; it doesn’t replace reading it.
Log the questions the system couldn’t answer. That list tells you which documents are missing, which ones converted badly and what your techs need to know that no manual covers. It’s often the most telling report the system produces.
Before rolling out to the whole team, run a pilot. Pick one equipment line, have your senior techs write twenty or thirty real questions they already know the answers to, and check each answer and its source. Fix what comes back wrong, then hand it to a few techs in the field for two weeks and ask them what they’d change.
What it costs
How Lithium builds it.
Lithium builds manual search systems as custom AI systems . Each one is scoped after a free discovery call and quoted before the build starts; most custom builds take one to two weeks. The price depends on how many documents you have, what shape they’re in, who needs access and where your techs will use it.
Bring a sample of your manuals, a list of questions your techs ask every week and a sense of which devices they carry. The discovery call covers whether manual search is the right fit or whether a simpler fix, like a cleaned-up shared folder, gets you most of the way.
Manual search pairs well with other field workflows. Techs who look up a spec on site can also turn their job notes into paperwork; see how voice notes become estimates and invoices . For terms like RAG and text recognition, the marketing glossary has short definitions.
