DatumOS
Turn any tender into a priced, send-ready quote — in minutes.
GAEB, PDF, Excel, or Word: DatumOS reads every position, matches it to your catalog, and prices it by your rules. Underneath sits the system of record for construction’s commercial data — one datum every quantity, price, and decision traces back to, from first estimate to final invoice.
The problem
Construction runs on files that don’t agree with each other.
A single project produces thousands of commercial data points — tender positions, bids, awards, change orders, invoices. Today they live in GAEB files, PDFs, spreadsheets, and inboxes, retyped at every handover.
Every handover is a retype
Tenders arrive as GAEB, PDF, Excel, or Word. Estimators re-enter them line by line into their own systems — days of skilled work per bid, spent on transcription instead of judgment.
Numbers drift between phases
The quantity in the tender, the price in the bid, and the amount on the invoice should agree. Spread across disconnected tools and formats, they quietly stop agreeing — and nobody notices until it costs money.
Knowledge is trapped in prose
What is actually being built — dimensions, materials, performance classes — is locked inside long-text specifications that no system can search, compare, or price automatically.
The platform
One datum. Every phase measured from it.
DatumOS is the operating system of record for construction, built in layers: an open data foundation that makes commercial construction data precise and machine-readable, workflow applications that turn it into finished work, and AI that reads what only humans could read before.
The record — pyGAEB
An open-source engine that reads every GAEB DA XML version and exchange phase into one canonical, decimal-precise, fully typed model. Parse, validate, compare, convert, write back. This is the datum itself — and it’s MIT-licensed, so you can audit every line.
Explore pyGAEB →The workflows — the DatumOS platform
One workspace built on the record, at app.datumos.build. The first workflow: tender-to-quote. Upload a bill of quantities in any format and get back a priced, branded, send-ready quote — in minutes instead of days.
Explore the platform →The intelligence
AI that classifies every position by trade, extracts structured specifications from long text, and matches items to product catalogs — always with confidence scores, and always with a human review step on every uncertain call.
The commercial lifecycle, one data model
The pyGAEB record layer already speaks every phase of the GAEB exchange — so every workflow we build starts from the same benchmark.
Products
Two products on one foundation.
DatumOS Platform
From tender to quote in minutes, not days.
- Upload any tender — GAEB, PDF, Excel, or Word. You never think about file formats.
- AI extracts every position: quantities, dimensions, materials, performance classes.
- Multi-signal matching against your product catalog with confidence scoring.
- Only uncertain lines ask for your judgment — everything else flows through.
- A branded, send-ready quote at the end. Developed with door and window manufacturers, built for every trade.
pyGAEB
The GAEB engine for the modern stack.
- Every GAEB DA XML version (2.0–3.3), auto-detected — plus legacy GAEB 90.
- All exchange phases: tendering, awarding, billing, quantities, costing, ordering.
- Validation, document diff with financial impact, version conversion, Excel export.
- Optional LLM classification — any provider, including fully local models.
$ pip install pyGAEB Explore pyGAEB → Why we start in Germany
GAEB-native from day one.
German-speaking construction already has what most of the world lacks: a decades-old standard for structured tender exchange. We build on it natively — and carry the same discipline to markets that still run on PDF.
Deterministic where possible, AI where necessary
Structured GAEB files are parsed exactly, not guessed at. AI is reserved for the genuinely unstructured — PDFs, prose specifications, catalog matching — and every AI call carries a confidence score.
Built for real German practice
Leistungsverzeichnis, Nachtrag, Preisspiegel, Aufmaß — the platform models the actual VOB/GAEB workflow, including alternative and contingency positions, framework contracts, and change-order traceability.
An open core you can audit
The record layer is open source under MIT. Your commercial data model shouldn’t be a black box — inspect it, extend it, run it air-gapped with local models if your policies require it.
Work with us
We’re onboarding design partners.
If your team receives GAEB or PDF tenders and prices them by hand — whatever your trade — we’d like to build with you. Pilots are free, run four to six weeks on your real tenders, and an NDA is standard.
Get in touch →Or write to us directly: hello@datumos.build