AI Takeoff: material quantities from plans, specs and aerial imagery, with a source on every line.
Upload the bid set. Keelstone reads every sheet, legend, note and spec section, extracts quantities across 25 trades, flags what the architect left inconsistent, and processes each addendum into a variance report. Review it yourself in minutes, or have an estimator QA pass before it reaches your estimate.
AI only. Quantities, discrepancies and the variance report are ready in minutes; your estimator verifies, edits or rejects each line. Every correction teaches the next run.
AI plus a human QA pass. The run lands in a review queue with a 48-hour promise; nothing is approved until every line has been checked, and accuracy after review is tracked per run.
Bring your own Anthropic, OpenAI or Google key. Every AI call is logged with tokens and cost per project; documents stay in your tenant behind row-level security.
What it does
PDF plans, specifications, legends, notes and addenda. Every sheet is split, text-extracted and OCR'd first, so nothing is skipped because it was scanned.
HVAC, plumbing, electrical, steel, concrete, masonry, civil, utilities, roofing, drywall, flooring, ceilings, painting, doors and glazing, fire protection, refrigeration, low voltage, equipment, demolition, landscaping, paving, snow and more.
Each line carries the sheet, page or spec section it came from, the exact text it was read from, and a confidence score. Nothing enters an estimate unverified.
Sheets in the index but missing from the set, references to sheets that do not exist, conflicting square footages, missing spec sections, addenda whose revised sheets never arrived. One click drafts the RFI.
Calibrate from two points or the printed scale, then take areas with deductions, lengths and counts straight on the drawing. Measured quantities join the takeoff as verified lines.
Landscaping, paving, snow and roofing from satellite or drone imagery: fetch by address, measure at true ground resolution, or let the vision model estimate surfaces for your review.
An assistant that answers spec lookups, scope cross-references and inconsistency questions from your project's own files, always citing the sheet or section.
Export with live formulas, one tab per trade, extended costs and totals. Addenda produce a variance report: added, removed and changed quantities against the previous run.
How a run works
- 1. Read
Current sheets, spec sections and addenda are gathered from the job folder. Rules find the stated quantities first.
- 2. Extract
Each sheet goes through the model with its matching spec sections and your past corrections. Optional vision pass counts legend symbols.
- 3. Check
Plan QA runs rules and an AI review: missing sheets, dangling references, area conflicts, spec gaps, addendum conflicts, scope-shifting notes.
- 4. Review
Verify, edit or reject per line or in bulk, draft RFIs, compare runs, export to Excel. Corrections become examples for the next run.
From our demo set
One run on an 11-sheet retail remodel set with a spec book and one addendum, using Claude Fable 5.1:
- Caught that Addendum 1 reissued A-101 while only Revision 0 was in the set
- Found a phasing plan and an equipment schedule referenced but never issued
- Reported the lay-in ceiling, storefront and painting scopes had no spec sections
- Answered the bid due date from the addendum, not the superseded ITB, with sources
Trades covered
AI usage is metered against your plan's allowance with your own provider keys. Self-host or run on Keelstone cloud.