The motion
One engine · Their documents · Tenant wall
One Corpusmith AI engine for every customer. Personalization is their documents behind a tenant wall — not a forked app. Bring one Civil / AEC project: specs, RFIs, submittals, SOPs, as-builts.
Book a demoThe motion
One Corpusmith AI engine for every customer. Personalization is their documents behind a tenant wall — not a forked app.
Find the painful workflow (RFIs, specs, SOPs) that eats the civil firm’s time. One job that already hurts — not a vague “make us more AI.”
Pull their documents into a private tenant. Same Corpusmith engine for every customer; their files only — not a forked app.
Chunk, embed, and index until status hits ready. Ready means ready — we do not pretend the corpus is live before it is.
Team asks in plain language; answers cite the source docs. Receipts are the product — a fake citations UI is not.
The product
One Corpusmith AI engine for every customer. Personalization is their documents behind a tenant wall — not a forked app. Civil Engineering first. Civil / AEC documents.
01
FoundationPostgres. Ingest Excel, CSV, and PDFs. Classify by department and project. Extract metadata and sections into structured tables so agents can filter.
02
BrainAn LLM on that private data — with a path to company-owned or local models later. Agents, skills, and tools sit on the same engine.
03
WorkDeterministic recurring work with a human sign-off. Drafts for people to approve. We do not silently file, send, or ship.
Built toward enterprise security and on-prem options. This is not a SOC 2 certification claim.
Civil / AEC
These are the shapes of a first project — not customer names, not case studies, not metrics.
Search every spec section and submittal in seconds; answers cite the page.
Learn more →Find prior RFIs and responses across projects so the team doesn’t reinvent answers.
Learn more →Turn retiring experts’ SOPs into a private brain the whole firm can ask.
Learn more →Knowledge layers
You get both — their corpus, locked behind hard tenant walls.
It's your data. You own the files, the index, and the answers.
Book a demo →Example Co. assistant
Illustrative tenant workspace
Ask · receipts later
Illustrative — not a citations UI
Sample specs, RFIs, submittals, SOPs. Citations will point at real files later.
Illustrative Civil / AEC workspace — sample docs, not a live corpus or a citations UI.
Book a demo →The company corpus — their tenant
Answers with receipts
Illustrative only. Citations will point at real files and chunks later — this page is not a citations UI, and it is not a case study.
What was approved on the waterproofing submittal?
The submittal log and spec section — not a guess. Illustrative, not a case study.
Which RFI changed this detail?
The RFI that closed, with the drawing or spec it touched.
What does the SOP require before we pour?
The procedure as written. A human still signs off.
Where is the as-built for that run?
The closeout set, not last week’s memory.
Humans in the loop
Drafts for people to approve. We do not silently file, send, or ship work. Receipts are the product; a fake citations UI is not.
Plain-language questions against their corpus. Receipts are the promise; citations to real files and chunks come later.
Pull together what the documents already say — the spec path, the RFI gap, the submittal comparison.
Drafts for people to approve. A human signs off — we do not silently file, send, or ship work.
What “good” looks like
The team can check the spec, RFI, or SOP the answer came from. Receipts, not vibes. No invented metrics here.
The next person on the job is not starting from a shared drive and a memory.
When the documents change, the private brain refreshes. A stale index is a failed engagement.
Start with the work
One company, one private Civil / AEC brain. A real engagement starts with one project, one painful workflow, and a success metric you can check.