A private brain, not ChatGPT with a PDF
The moat is a company model plus receipts, not a nicer prompt box. Hard tenant walls. Originals stay the source of truth. Their environment and data never train shared models.
About
A private company brain for Civil Engineering.
Corpusmith builds a private company brain from the documents you already have. One Corpusmith AI engine for every customer — personalization is their documents behind hard tenant walls, not a forked codebase per client. Upload specs, RFIs, submittals, SOPs, as-builts, contracts, wikis, and PDFs; we turn them into a searchable knowledge base that only that company can use.
Your team asks in plain language and gets answers grounded in your files — not the public internet, and never mixed with another customer’s data. We start with one painful workflow and one tenant, prove it, then expand. Humans stay in the loop: drafts and answers for review, not silent automation.
First beachhead is Civil Engineering (Civil / AEC documents). Environmental later. The system is Collide-shaped for work that runs on trapped documents — we are not an oil-and-gas product.
Why this exists
The knowledge that actually runs a job is trapped in specs, RFIs, submittals, SOPs, as-builts, and people’s heads — and it walks out the door when those people leave. Corpusmith takes the documents a Civil Engineering company already has and turns the same AI engine into their private brain — hard tenant walls, not a forked codebase. Civil / AEC documents first. Environmental is a later beachhead. We stay out of oil and gas.
The moat is a company model plus receipts, not a nicer prompt box. Hard tenant walls. Originals stay the source of truth. Humans sign off on drafts. One real company, end to end, before multi-customer admin. Their environment and data never train shared models.
Receipts now. Humans sign off.
Ask and analyze against the corpus. Generate drafts later — a person approves. Citations will point at real files and chunks later; we do not invent that UI on the brochure.
Civil Engineering first. One tenant before a platform.
One real company, end to end, before multi-customer admin. Specs, RFIs, submittals, SOPs, as-builts — Civil / AEC documents. Environmental is later. We stay out of oil and gas.
What this is not
We are not building a community app.
Paid add-ons (calls, SMS, generic websites) stay off the landing page until a client asks.
If a feature is not in how it works, we do not ship it for the brochure.
- A community app, forum, or marketplace of AIs
- ChatGPT with a PDF, or a nicer prompt box over the public web
- An oil-and-gas product, or a fake citations UI
- A multi-customer admin console before one real company is proven end to end
- Silent automation — drafts need a human sign-off
- Fake metrics, logos, or invented case studies
Where the site is
Brochure is live. Vertical is Civil / AEC.
The public site is a brochure for the north star. Marketing pages stay runnable without keys. Citations will point at real files later — this is not a citations product yet.
Near-term website work lives on the site plan. Product scope stays in docs/PRODUCT.md.
Start with the work
Find your first workflow.
Discover, ingest, private brain, ask with receipts — then expand. One Corpusmith AI engine. Their documents. Hard tenant walls.