Architrave

Your data

How We Handle Your Data

Your estimates, sub numbers and job costs are the most sensitive files in your company. We built how we work around that, not around whatever AI tool is popular this month.

The short version

  • Raw files never touch public AI.
  • About 90% of the work is deterministic code, not AI.
  • Sensitive text is handled by local open-weight models.
  • Cloud models only ever see scrubbed data.
  • If you want, everything runs on your own infrastructure.

Remote work

When you send us files, they arrive over an encrypted connection and are processed on our own secured hardware. Not a shared cloud workspace, and not a public chatbot. Raw files never touch public AI.

On-premises work

For clients who don't want files to leave the building at all, everything runs on your infrastructure. We spec the hardware to run the models, set it up and test it on site. Your data never leaves your network.

Either way: how the work actually gets done

About 90% is deterministic code. Most of what we do doesn't need AI at all. Checking quantities against a takeoff, footing totals, comparing one revision to another and lining up sub proposals is done with plain code (local Python) that gives the same answer every time and can be checked line by line. That's more reliable than asking a model to guess, and none of it leaves our hardware or yours.

Sensitive text is handled by local open-weight models. When work does need AI, such as reading proposal exclusions, spec language or daily reports, it runs on open-weight models on our hardware or yours. Those models don't send anything to an outside company.

Cloud models only ever see scrubbed data. On the rare occasion a task benefits from a large cloud model, it only gets data that's been stripped of company names, client names, project details, pricing and anything else that identifies you.

An estimator checks every finding. Automated checks surface issues. A person who has built estimates for a living verifies each one before it reaches you.

Model-agnostic by design

We don't sell or depend on any one AI company's product. Our setup is model-agnostic, so we use whatever model does the job best and can switch when something better or safer comes along. You're never locked into a vendor because of us.

Your files, your call

  • We're glad to sign your NDA before you send anything.
  • We only use your files for the work you hired us to do. We never use them to train models or share them with anyone.
  • When the work is done, we delete your files on request, or keep them only as long as you want us to for repeat work.

Questions we hear

Do you use ChatGPT on our bids?

Not on raw files, ever. Cloud models only see scrubbed data, and most of the work doesn't use AI at all.

Can our IT team review the setup?

Yes. We're happy to walk your IT team or outside provider through exactly how files move and where they're processed.

What does on-premises cost?

On-premises Custom Tools & Private AI engagements start at $7,500 plus travel, with hardware purchased separately. See Custom Tools & Private AI.

The estimator still makes every call. We just make sure nothing goes unread.