The manual is unsearchable
A 400-page scanned service manual on a phone is not a reference. Technicians call the office instead, and the office calls the one person who knows.
For field service companies
Your technicians already have the answer — it is on page 118 of a manual nobody can find on a roof in August. Tinker Assist searches your company’s manuals, your approved fixes, and the documents attached to the job, and answers with the page it came from.
What does code 33 mean on the Carrier 58MVC, and how do I clear it?
Code 33 is a limit circuit lockout. The furnace has opened the limit switch during a heating cycle and locked out to protect the heat exchanger.
Safety: disconnect power at the service switch before removing the blower door.
Source: Carrier 58MVC Furnace Service Manual, page 118 — opens at that page.
The problem
Every field service company already owns the answer to almost every call. It is distributed across a few thousand PDFs, a shared drive nobody has reorganised since 2019, and three people who have been there long enough to remember.
A 400-page scanned service manual on a phone is not a reference. Technicians call the office instead, and the office calls the one person who knows.
The thing that actually worked at 2am gets typed into a work order nobody reads again. The next person on the same equipment starts over.
Site drawings, submittals, and as-builts live in a Drive folder per job. They are the most relevant documents on the call, and the hardest to reach from a ladder.
What Tinker does
Every answer grounded in your documents names the manual and the page, and opens the original PDF at that page. Nothing has to be taken on trust.
Figures and wiring diagrams on the cited page are extracted during processing and shown alongside the answer, so a technician sees the drawing, not a description of it.
Your team writes up what worked. Management approves it. From then on it answers questions alongside the manufacturer’s manual, and it is cited the same way.
Attach a job’s drawings, submittals, and synced folders to the job. Questions asked on that job search its own documents first, and never another job’s.
For a recall, a bulletin, or a model your library does not cover, Tinker can consult the open web — preferring manufacturer sources, labelling them clearly, and never presenting them as your documentation.
If your documentation does not support an answer, Tinker says so plainly rather than filling the gap. A confident wrong answer on live equipment is worse than no answer.
Built for the work
Tinker Assist is an AI assistant for field service technicians. It searches your company’s equipment manuals, your management-approved fixes, and the documentation attached to a specific job, then answers a technician’s question with the page it came from.
A general chatbot has never read your manuals. Tinker answers from documents your company uploaded, and every claim it makes from them carries a citation naming the document and the page. When your documentation does not cover the question, Tinker says so instead of inventing an answer.
PDF manuals: service manuals, installation guides, wiring diagrams, parts catalogues, and job paperwork. Tinker extracts the text page by page, keeps the page numbers, and pulls out the figures on each page so a diagram can be shown next to the answer.
No. A technician opens the app, picks the job they are on, and types the question they would have asked a senior tech. The answer comes back as a direct answer, ordered steps, tools and parts, and any safety warnings the manual attaches to that procedure.
Yes. A technician can write up a fix — the problem, what actually resolved it, and the steps — and submit it. Management reviews it, and only an approved fix is ever used to answer anyone else’s question.
Yes. Tinker installs to a phone, tablet, or desktop from the browser, runs full-screen like an app, and is built for one-handed use. It needs a connection to answer a question, because the answer comes from your documents on the server.
We will set up a pilot with your documentation, your technicians, and your equipment — so you are judging Tinker on your own work, not on a demo dataset.