Software for general contractors, run by an AI crew.

AI for general contractors is not a chatbot bolted onto a construction logo. It is a set of agents put on the measurable work a GC actually does: reading the set, pulling scope, measuring quantities, pricing on your history, leveling sub bids, reconciling contracts, tracking submittals and RFIs, assembling pay applications and the closeout packet. Each agent works inside your process and your templates, explains every number it produces, and stops at the human gates your team already uses. That is the difference between an assistant that answers questions and a platform that does the job.

The economics start with throughput. In our discovery calls, GCs describe the same ceiling: bid volume is capped by estimating hours, and win rate is capped by how carefully each bid can be checked before it goes out. When scope, takeoff, and leveling stop consuming the week, the same team prices more work, checks each bid harder, and spends the reclaimed time on site visits, sub relationships, and strategy. More bids at higher quality is how revenue moves, and it moves without adding headcount.

The second problem this solves is quieter: the senior estimator bottleneck. Labor multipliers for height and pitch, how facade variation changes coverage, which subs pad their allowances. That judgment lives in one head, and every new estimator and every new office starts from zero. Ruh writes the assumptions down, keeps the corrections, and prices the next bid on what your best estimator taught it. The knowledge stays in the company when the person is on vacation, on site, or retired.

Scope automation begins with the package as it came in: the drawing set, the specification book, addenda, and the estimator's own scope notes. The set is read sheet by sheet and classified by discipline and level, so a building level is counted once even when it is drawn on five sheets. The specification derives the trade-to-division map, so the estimate's structure follows the spec book rather than a template. Each scope item is tied to the sheet, photo, or document it was read from and carries a state: measured, flagged for review, or assumed. Where the documents do not support an item, a formal RFI is raised with a recipient and a priority, and the run halts at that gap rather than pricing past it.

The takeoff engine works on the drawing PDFs and CAD files directly. It classifies sheets, calibrates scale, measures room areas and wall runs, counts doors, fixtures, and symbols, and reconciles vector and vision readings against each other. Every quantity carries a verification state and a source, and every trade pulls from its own physical basis: flooring from floor area, ceilings from ceiling area, painting from wall area, doors from counts. A floor area is never reused as a wall area. The run reports its page coverage, so a partial run finishes honestly as partial with the unmeasured pages named.

Pricing comes from your price book, your assemblies, and your closed jobs, not a generic database. Each line carries its pricing source, most grounded first: a firm sub quote, your own unit cost, a derived assembly, or a labeled allowance. The estimate is assembled with your general conditions, fee, contingency, and bond cascade, then exported into your real estimating workbook and proposal template by template parity, not a lookalike. Anything the system could not confidently place is listed in a report rather than hidden in a total.

A GC compares 30 to 40 trades with 4 to 7 bids each, around 300 proposals per project, against its own estimate. Ruh parses each received quote from its document, keeps the highest revision, and levels with rules you can read. Only the lowest real quote per trade is used. A firm quote replaces the estimate line rather than stacking on it, so scope is never double-counted. A combined multi-trade bid is carried unplaced rather than mis-leveled as one trade. Gaps are priced as labeled allowances, never as silent zeros, and a coverage score tells you how much of the bid is hard-priced before you submit.

Bid invitations are logged by trade and channel, and follow-up keeps running without an estimator spending a day on fifty phone calls. The drafts are yours to approve; your team owns the relationships. After award, the same record tracks each sub's schedule adherence, RFI volume, defect rate, claim history, and payment timeliness, and scores them on every new bid package: who to invite, who to deprioritize, what to ask in prequal for new entrants, with the reasoning attached.

When the bid is won, the approved estimate freezes as the as-sold baseline. The execution budget compares original, approved changes, revised, committed and buyout variance per row against that baseline, so a subcontract awarded above what was carried is visible as a buyout overrun before any approved change. Spending, payments and reconciliation stay in your accounting system. The baseline schedule is generated from the takeoff, with durations derived from measured quantities and production rates and milestones chained on working days. As the job runs, recorded impacts move the projected finish and each milestone is flagged on track, at risk, behind, or complete.

Contract terms are reconciled against the proposal, including qualifications and exclusions, so what is signed matches what was priced. Each executed subcontract is broken into a schedule of values that must sum to the commitment, the lines a sub bills against. A change event is captured from a sub, a drawing revision, or the field, assessed for validity and pricing, attributed to a budget trade, and carried to the owner's decision. Approved amounts flow into the revised budget and the next pay application; declined ones stay on the record with the reason.

Submittals move through a fixed review pipeline with a business-day clock on each review stage and a stamping gate before closeout. RFIs carry ball in court, an answer on the record, and a link back to the estimating stage that raised them when they began as pre-bid questions. Daily logs are versioned manpower records that can arrive from the sub portal or by email. Punch items pin to a drawing or a GPS point with photos and can link to the commitment responsible for the fix.

A pay application is one billing period against a commitment's schedule of values, with retainage and compliance gates such as certified payroll where the job requires it. Invoices are matched to commitments and flagged before money moves. Closeout is not a new data model: the handover packet is a view over the documents, the estimate and its versions, buyouts, allowances, the baseline and the schedule, open RFIs and conflicts, and the stamped-submittal bundle. Nothing is re-keyed at the end of the job.

Ruh wraps the stack a GC already runs rather than replacing it: Procore for the project record, Bluebeam for markups and takeoff exchange, Sage and QuickBooks for the ledger, Excel for the workbook your estimators live in. Exports land in your own templates. The agents use those tools the way a new hire would, so nothing has to be migrated before the first bid runs.

The objections we hear are fair, and they have direct answers. Data ownership: Ruh runs inside your own tenant, on your own AI license, with no training on your data, and client-controlled hosting is supported. Cost: there is no per-takeoff token bill that balloons on a big set; the build is scoped once, then a flat monthly plan. Accuracy: it is earned, not claimed. We run your real drawings against estimates your team already trusts, your corrections teach the system, and a human signs every line. Start with a free takeoff and judge it on your own work.

Adoption inside a GC is its own project, and we build for that too. The pattern we see across firms, including private-equity-owned groups pushing AI hard, is one champion with no internal peers, an IT department speaking a different language, and management that needs a sponsor before any process change moves. So the rollout is designed around that reality: one real bid, run live, with results your president can read. The walkthrough produces the internal case, not just the takeoff.