AI for architecture practices.
Governed, practical, built from inside practice.

We have built a governed integration that lets Claude work inside Revit (batch sheets, shared parameters, view templates, family audits) with a ledger of every action and a check on which model is open before anything changes. Proven routines are promoted into reusable skills, so good practice stops living in one person's head.

Regulations lookups grounded in the Approved Documents, and multi-agent pre-checks that split a scheme description across the relevant parts and flag where requirements pull against each other. They support review; they don't replace it.

Fee proposals, specifications, drawing registers, meeting notes and bid content generated from your own data, then checked by your team.

Strategy drawings, issued same day

A layout change raised a compliance question that needed design-intent drawings before the package could move. Working inside the live Revit model, the session produced an explanatory note with sketches drawn from real model coordinates, a new strategy sheet and a revised drawing, assembled with the issue sheet - drawn, checked and issued the same day, with the specialist consultant holding the final design.

A drawing issue that normally takes days, done in one session.

Grey model to presentation CGI in under a minute

Render engines need fully dressed models; live models never are. This pipeline reads the Revit view and its annotations - ceiling plans, finishes legends - and drives image models directly. Every render is automatically scored against the real Revit geometry before anyone sees it.

Hours of model dressing replaced by a checked, sub-minute render.

Hundreds of fixtures placed in Revit from a CAD layout

Fit-out layouts arrive as CAD drawings carrying hundreds of fixture blocks, traditionally matched to families and placed by hand. One session built the process end to end: the families built, every instance placed and marked to its position, with a register and an update route for the next drawing revision.

A day of error-prone placement became a supervised automated pass.

The same process, dropped onto the next project

Automation only counts if it works on a drawing it has never seen. Re-run on a different layout, the process placed the fixtures without manual Revit input, with the genuine judgement calls - unknown blocks, size mismatches, bespoke joinery - flagged for human review rather than guessed at.

A one-off trick became a reusable practice tool.

A 2D drawing became a data-rich Revit family

A furniture item existed only as a dimensioned 2D general arrangement drawing. The AI read the geometry from the CAD file and wrote code against the Revit API to build the family, loaded with real data: materials, part references, weight, fixing notes. Its first attempt read part of the frame wrong - the review step caught it, and the corrected family went in the model. That check is the process working, not failing.

Schedule-ready families from supplier drawings in minutes, review built in.

Practice AI policy and assistant

A UK architecture practice

An in-house AI site with a formal policy, a do's and don'ts grid, an Ask-the-Policy chat, a prompt library and a policy-governed assistant that enforces confidentiality and flags Building Regulations, fire, structural and CDM risk.