Swapp

SWAPP sits in the least glamorous corner of architectural practice — construction documentation — and that is exactly why it matters. Dims, tags, views, sheets: the work that consumes the majority of production hours on a real project and wins nobody a design award. In my look at the AEC automation space, most AI tools chase schematic design renders or feasibility studies; SWAPP goes after the CD set itself, running inside Revit and ArchiCAD rather than around them. It is built for architecture firms with real production volume, not for a solo designer who wants a pretty concept model.

Positioning first, because it shapes everything else. SWAPP sells itself as an operational intelligence layer rather than a plugin. The distinction is not marketing fluff: the platform trains on a given firm's own production history, annotation rules, and QA logic, then applies those conventions across projects. A dimension standard your senior staff enforce by eye becomes something the software enforces by rule. The vendor claims up to 80% of construction documentation tasks are automated — an unverified vendor figure, but the direction matches what firms report in SWAPP's published testimonials (MYS Architects claims an eightfold reduction in manual workload; that too is a vendor-published quote). The company behind it has real pedigree: co-founders Eitan Tsarfati, Adi Shavit, and Noam Gat came out of Homestyler and Autodesk Israel, and the firm has raised roughly $18.5 million across a 2021 seed and a 2023 Series A, per investment trackers. It has been shipping since 2019 — seven years on one problem.

What the product actually does: three strands. Documentation generates complete CD sets — dimensioning, tagging for doors, walls, windows, fixtures and casework, view and sheet creation — grounded in the firm's standards. Modeling covers 2D-to-3D conversion, finish schedules, parameter work, and validations from SD through CD. QA/QC and compliance includes agentic ADA checking that both finds accessibility issues and proposes fixes in the model. Over the top sits "Frank," an agentic assistant that operates inside the live Revit (and ArchiCAD) model — you trigger it conversationally or accept its proactive suggestions, you approve, and it executes: modeling floors, placing dimensions, resolving collisions, running checks. The homepage headline statistic is 35M+ square feet of production-grade documents delivered and 104,300+ production hours saved over two years (vendor figures). Enterprise posture is unusually concrete for this sector: ISO/IEC 27001:2022 certified (cert #1125599), hard tenant isolation, native BIM files never sent to third-party LLM providers, and a stated commitment that customer data never trains general models.

My testing notes. I worked through SWAPP's published sample CD sheets and a walkthrough of a Frank session rather than a full firm deployment — this is enterprise software sold by demo, so hands-on time is necessarily bounded. Two observations stood out. First, the annotation behavior is the real story: in the walkthrough, Frank dimensioned and tagged an enlarged plan and then did a legibility pass, detecting overlapping tags and re-placing them against the firm's graphic standard. That second step — cleanup as a first-class operation — is what separates this from Revit macros that spray dimensions and leave a human to tidy up. Every automation tool I have used in this space produces output that needs a cleanup pass; folding the cleanup into the run is the difference between saving time and moving it. Second, the "workflow memory" idea behaves like institutional capture rather than generic machine learning: the system ingests the firm's standards documents and production history, and the intelligence stays firm-owned. For practices where two or three senior people hold decades of drafting judgment in their heads, that reframes the product as succession insurance — the pitch SWAPP itself makes, but one that matched what I saw in how standards are stored and applied per firm. I also priced out the self-serve tiers that appeared on SWAPP's pricing-v2 page during my check: a free Starter (1,000 credits/month), Pro at $45 per user per month (10,000 credits), and Team at $95 per user per month (25,000 credits), all behind a waitlist or early-access request at time of writing (source: swapp.ai/pricing-v2). The main pricing page, meanwhile, still describes a purely credit-based, custom-quote model — two pricing stories on one site, which I read as a product mid-transition.

Limits, and there are meaningful ones. SWAPP does not design for you. It does not replace design judgment, produce concept options (its earlier product life, aimed at developers, did algorithmic plan options; the current product is production-first), or handle interiors-specific detailing beyond what its modeling strand covers. It cannot be meaningfully trialed by a solo architect — the value premise is that the AI learns a firm's standards, which requires a firm with standards, production history, and enough Revit/ArchiCAD volume to amortize onboarding. If you produce a handful of small projects a year, this is a fire hose. The self-serve tiers being waitlist-gated also means you cannot simply sign up and start; enterprise engagements run through a demo and custom quote, and the credit costs per task type are not published in units a buyer can model in advance. Third-party price trackers put sheet-based pricing historically around $1/sheet with custom quotes by project size, but I could not verify that against a current vendor page, so treat it as unconfirmed. Reliability of the AI output also deserves scrutiny in any pilot: documentation errors in a CD set become construction errors, and while the QA/QC strand exists precisely to catch them, a firm should still run SWAPP output through its normal check process during evaluation rather than trusting the "up to 80%" figure.

Ecosystem and support: dedicated account management at enterprise tier, priority support at Team (24h) and email support (48h) at Pro, per the pricing-v2 page. Named customers include Page, VLK Architects, SNHA (a Woolpert company), MOREgroup, HGA, and Huckabee — logos that suggest real institutional adoption rather than a pilot graveyard. On reputation signals: ISO certification, published security page, named exec team, and venture backing all check out; user-review volume on public aggregators is thin, which is normal for enterprise AEC software sold by relationship.

One more thing on where SWAPP sits relative to the rest of the AI-for-architecture field. I have tested a fair number of tools in this category recently, and the split is fairly clean: generative design tools produce options at the front end, render tools produce imagery for clients, and document-intelligence tools read issued sets after the fact. SWAPP is one of the very few attacking authoring-time production in the live model — the moment where a drafter would otherwise be clicking dimensions for the sixth hour of the day. Its QA/QC and ADA strand extends that into checking work: agentic accessibility review that flags issues in the model and proposes fixes is the kind of task that usually lives in a checklist nobody finishes. Whether the ADA checking meets a firm's legal compliance standard is a question for counsel, not for a vendor demo; treat it as a first-pass catcher, not a code consultant. On onboarding, expect a real commitment: the platform ingests your standards and production history before it can enforce anything, and the firms quoting big time savings are a year or more into that relationship. Budget evaluation time accordingly.

Verdict: rated 4.1 — the rubric weights core functionality depth (40%) heavily, and SWAPP genuinely automates the deep, standards-bound production work its category competitors avoid, backed by value that firms can measure in hours, credible security posture (supporting the reliability and ecosystem legs), with deductions for opaque enterprise pricing, waitlist-gated self-serve, and the near-impossibility of a low-commitment trial. If you run a mid-to-large practice drowning in documentation, book the demo and demand a pilot on a real project with your standards loaded. If you are small or design-led first, this is not your tool yet.

Sources

• https://www.swapp.ai/
• https://swapp.ai/pricing/
• https://swapp.ai/pricing-v2/
• https://swapp.ai/products/documentation
• https://www.swapp.ai/products/frank/
• https://www.swapp.ai/security/
• https://www.swapp.ai/who-we-are/
• https://indexed.vc/companies/swapp
• https://startupintros.com/orgs/swapp
• https://venturebeat.com/ai/swapp-raises-7-million-to-automate-construction-planning-with-ai
• https://londontechwatch.com/2021/02/swapp-construction-planning-ai-platform-eitan-tsarfati
• https://prnewswire.com/news-releases/amar-hanspal-joins-swapp-advisory-board-and-invests-in-their-growth-301563978.html
• https://resources.rework.com/tools/ai-tools/best-ai-tools-for-architects-2026
• https://www.aecplustech.com/blog/swapp-the-ai-co-pilot-for-construction-documentation