Your drawings stay yours.
How your files are handled, step by step. Not a badge: the actual mechanics, and what we don't claim yet.
How your data is kept apart
- Each workspace's data is isolated at the database level (Postgres row-level security), not only by application code.
- Uploads and reports are stored under your workspace's own storage prefix, never shared across workspaces.
- Drawing pages are read by Anthropic's Claude API for text and data. The structural code checks (AISC, ACI, ASCE, NDS, TMS) run on our own servers, with no AI involved.
- Reports are built on the server and redacted for unpaid views. The full finding text never reaches a browser that hasn't unlocked it.
- Every action a platform staff member takes is written to an append-only audit log.
What we check, and from where
New York City. Every check cites a named NYC requirement: the DOB Plan Examiner Guidelines, the DOB Technical Reports list (TR-1 special inspections, TR8 energy items, Tenant Protection Plan), DOB's Enforcement Action Bulletins, and the 2022 NYC Building Code, all chapters, from nyc.gov.
New Jersey. Checks cite the Uniform Construction Code regulations, N.J.A.C. 5:23, from the Department of Community Affairs: what a permit application must contain (5:23-2.15), New Jersey's amendments to the IBC (5:23-3.14) and the Rehabilitation Subcode for existing buildings (5:23-6). The IBC text itself belongs to the ICC; we link to its free read-only edition rather than copy it.
Checks that follow fixed rules are labeled as such. Items read by AI are labeled "verify".
Not yet covered: the zoning math (we check that zoning information is stated and consistent), full energy-code compliance (we check the statement, the energy analysis and the code edition), and MEP schedules. This is advisory screening, not an engineering opinion or a plan review, and not for third-party reliance. The professional of record owns compliance.
Location, never assumed
The code a set is checked against depends on where the building is, so we never guess it. The project address is required. Before anything is read by AI, the drawings' own title blocks are compared with it, and a set that points somewhere else is rejected with the words that show it. After the AI read, the project's state is checked again before any score is given; if it disagrees, nothing is scored, your free check is returned and a refund is filed for any payment.
Retention you control
Your uploaded drawing set is scheduled for deletion after your workspace's retention window (90 days by default), shown on each filing's page next to a "Delete now" button that removes it immediately. Your Readiness Score and report stay available after the source file is deleted; only the original drawing set is removed.
Never used to train AI
We don't use your drawings or reports to train or improve any AI model, and there is no setting that turns it on. Our AI provider, Anthropic, doesn't train on API inputs or outputs under its commercial terms, and deletes them within 30 days except where its usage policy requires otherwise.
Sub-processors
Contact and audits
To report a security concern, contact the team through your normal point of contact during this pilot.
A SOC 2 Type I audit is planned as the product leaves the testing phase. We won't claim compliance we don't have; this page will say so when that changes.