Ship AI features contractors trust.

01 · Harness engineering

Agents that finish long construction tasks inside your product.

A takeoff across a full drawing set, an estimate or a contract review is hundreds of steps, not one prompt. We build harnesses that keep the model on a defined path, so long jobs finish correctly on every customer's project, not just the demo set.

• Built for long-horizon tasks: drawing takeoffs, estimates, contract and specification review.
• Full control of the run: defined steps, permitted tools, and validation at every stage.
• Checkpoints and resumable runs, so hours of work survive a failure instead of starting over.
• Evaluation suites built from real drawing sets, specs and contracts, so you measure accuracy before every release.
• Fits your stack: your backend, your queues, your APIs, and the integrations your customers already use.

02 · Memory layers

Memory that respects every customer, project and user.

AI is only as good as what it remembers, and who it is allowed to remember it for. In a multi-tenant product that is an infrastructure problem. We design memory in layers and engineer how information moves between them.

• Tenant isolation enforced in the data layer, not in the prompt. One customer's data never reaches another's.
• Inside each customer, information moves up only when someone promotes it, and down only to people cleared to see it.
• A private working note never surfaces in another team's estimate. A company-wide rate change reaches every live project.
• Mapped onto your existing auth and permission model, so there is one source of truth for who can see what.

03 · Cost engineering

Unit economics that leave you a margin.

Model prices differ by orders of magnitude. Across thousands of users and drawing sets, that difference decides whether an AI feature grows your gross margin or eats it. We measure cost per task and engineer it down.

• Routing: each step goes to the smallest model that does it reliably. Frontier models only where they earn it.
• Smaller specialised models, trained on construction work, for high-volume steps like sheet classification and extraction.
• Caching and batching for repeat work across drawing sets and revisions.
• Cost per takeoff, per estimate and per review, tracked per customer, so you can price AI tiers with confidence.

04 · Human in the loop

Review flows your customers will actually use.

In construction, who signs off matters as much as what was done. Your customers will only rely on AI output their own people can check and approve. We build that verification into your product, following their chain of authority.

• Role-based authority mapped to your permission model: each person sees and approves only what their role allows.
• Layered review: an estimator checks the takeoff, a PM approves the estimate, a director signs off the contract position.
• Low-confidence items flagged and routed for review, so people check what matters instead of everything.
• A full audit trail of every AI action and human decision, and corrections fed back into your evaluation suites.

05 · Tooling

Proven building blocks, so you don't start from a blank repo.

AI should not sit in a chat box next to the real work. We have built the pieces that put it inside the tools, running alongside people on the same drawing, document and record, and we bring them to your build.

• Takeoff Beam: our takeoff tool. The AI marks up the drawings and estimators verify them.
• Our own collaborative document editor, built on Tiptap, where AI and people draft proposals, reports and contract reviews together.
• Field capture by voice, photo or message, flowing straight into the office record.
• Connectors for the platforms contractors already run, such as Procore, SharePoint and Google Drive.