Home › AI assistant

Ask Photon

An assistant that answers from your procedures, not from the internet.

Staff ask a question in plain English and get an answer drawn from your own written operating procedures, your trade knowledge base, and the notes on the job in front of them. When something is not written down, it says so instead of inventing it.

Start free trial →7 days free · No credit card

Software nobody was trained on is software half the company avoids.

Every field service platform has screens most of the team never learns. The knowledge lives with two or three people, gets asked for by text message, and disappears entirely when those people are out. Training documents help right up until a screen gets reworded and the document quietly becomes wrong.

SolarProjeX ships a written manual and an in-app assistant that read from the same source. Over a hundred documented topics cover what each screen does and where it stops. The manual renders them filtered to the person reading — a technician's manual has no payroll pages in it — and the assistant answers questions from the same material.

The freshness check is mechanical, not aspirational. Each topic is fingerprinted against the actual labels on the screens it describes. Reword a button and the topic is flagged for review; rename it entirely and the assistant drops that topic rather than answering from a document that names a control which no longer exists. An assistant that confidently describes a button nobody can find is worse than no assistant.

  • Answers drawn from your own documented procedures, not general web knowledge
  • Also reads the trade knowledge base and the notes on the job being asked about
  • One source of truth behind both the written manual and the assistant
  • Manual filtered by role and access — people see the pages that apply to them
  • Topics fingerprinted against real screen labels, so stale documentation is flagged automatically
  • A topic naming a control that no longer exists is withdrawn rather than served
  • Per-person on and off switch, plus a company-level usage ceiling
  • Printable manual generated from the same corpus for onboarding and handouts

How it works

One written source, two ways to read it.

💬

Ask in plain English

"How do I split a reroof into two trips?" Answered from the procedure that describes it, on the screen you are already on.

📚

One corpus, no forks

The manual and the assistant read the same documented topics, so they can never drift into telling two different stories.

🔎

Knows the job in front of you

Questions about a specific job can draw on that job's own notes, not just the general procedure.

🚫

Silence beats a guess

If the answer is not written down, it says so. A confident wrong answer about your own process costs more than no answer.

Frequently asked questions

Does the assistant make things up?

It answers from your documented procedures and the job's own record. Where documentation is missing or has gone stale against the real screen, it withdraws that topic and says the answer is not written down rather than filling the gap.

Can we control who uses it?

Yes. It can be switched off per person, and there is a company-level ceiling on usage so office questions cannot consume an unbounded budget.

Does it carry our brand?

The manual and assistant text are written against brand tokens rather than hardcoded names, so a white-labelled deployment reads as your company's assistant, on your domain.

What keeps the documentation from going stale?

Each topic is fingerprinted against the literal labels on the screens it describes. Change a screen and the affected topics are flagged for review automatically, with the badly broken ones taken out of service until someone fixes them.

Everything SolarProjeX does, built for solar

One platform from the first lead to the last warranty claim. Every capability below is part of it.

Run your solar service on one platform.

Start free in 30 seconds. No credit card. Cancel anytime.

Start your free trial →