The Adaptive Construction Intelligence Platform

The all-seeing brain of construction.

Kynatix reads a construction drawing and returns an itemized estimate — line items, a phased schedule, a risk list, and the assumptions behind every number.

Not yet purchasable. The estimating pass is written and runnable from source; it has never been deployed to a host, and there is no sign-up.

01 · READ

It reads the drawing.

A general-purpose vision model reads one uploaded drawing — PDF, PNG or JPG, up to 3.5MB — and reports the drawing type, approximate area, stories, rooms, and the materials it can make out.

02 · PRICE

It prices what it read.

The same pass returns itemized line items with quantity, unit and unit cost, a phased schedule, a risk list, and the assumptions behind every number. Line totals are recomputed server-side and confidence is capped at 0.75 — you verify before you bid.

The problem

Construction software is siloed, static, and blind.

Billions in tech, yet the estimate is still built by hand — and nothing a project teaches you flows into the next bid.

01Manual takeoffs — slow, error-prone, brutal against a bid deadline.
02Static cost libraries ignore your real productivity, vendors, and pricing.
03No closed loop — field actuals and change orders never update the model.
04Late change orders — plan-vs-site gaps surface as costly disputes.
05Design/permitting disconnect — cost & code risk appear after drawings are done.
06Reactive decisions — historical reports instead of predictive signals.
What exists

One estimating pass — written, not shipped.

This is the whole of the built product today. Everything under Roadmap below is unbuilt, and the product app has never been deployed: it is excluded from the repository build script, and this repository contains no deployment pipeline.

one drawing → vision read → line items · schedule · risks · assumptions → server-recomputed totals

The drawing goes to a general-purpose vision model, not a construction-trained one. Every line total is recomputed server-side rather than trusted from the model, self-reported confidence is clamped, and each estimate is refused before it runs if it would exceed a hard per-estimate spend ceiling.

Roadmap

Tool → platform → intelligence layer.

Intent, not inventory. Only Phase 1 has code behind it, and that code is not deployed.

built · not deployed

Phase 1 · Single-drawing estimate

Vision read of one drawing into line items, schedule, risks and assumptions; multi-tenant projects and estimates; auth, billing and email wired but keyless.

roadmap · unbuilt

Phase 2 · Per-client brain

Tenant fine-tunes, dynamic cost/productivity libraries, forecasts, bid-win probability.

roadmap · unbuilt

Phase 3 · Design-build + XR

Concept-to-cost, prompt-to-detail CAD, VE, submittal/permitting AI, VR walkthrough.

roadmap · unbuilt

Phase 4 · Full brain

Closed-loop lifecycle, predictive CO/risk, agentic workflows, wearables & IoT.

Plans

Pricing isn't set yet.

There is no published price and nothing to buy. The product app has never been deployed and no checkout prices are configured, so any number here would be a number we invented. Access is by request while that's true.

Request access →

Kynatix is in active development and is not a purchasable product. The estimating pass described above exists in this repository but has never been deployed to a host, and everything under Roadmap is unbuilt. No accuracy figure, outcome number, or performance guarantee is claimed anywhere on this page, because none has been measured.