SuperNX / Security
Security & data handling
Your .prt files are your IP. This page states exactly what happens to an uploaded part — where it goes, who touches it, and when it is deleted. If a claim here ever drifts from what the service does, that is a bug: tell us and we fix one side or the other.
The short version
- Parts upload over TLS 1.3 to encrypted object storage; analysis runs server-side; the source .prt is deleted automatically the moment the analysis completes — not within a day, immediately.
- The outputs — report, toleranced model and rendered views of the part — are yours. They persist until you ask us to delete them.
- Model-derived data (extracted feature data and rendered views) goes to Anthropic for AI inference under commercial API terms: not used to train models. Nothing goes to any other third party.
- No ads, no trackers on your data, no sale of uploaded content, no training on customer parts — ours or anyone else's.
- On-prem deployment (pilot) runs the whole pipeline inside your infrastructure — nothing leaves your site.
What happens to your file
- Upload. The browser sends a single .prt file (≤200 MB) over HTTPS to object storage on Cloudflare.
- Analysis. A dedicated compute node downloads the file into an isolated working directory, reads the model, extracts geometry and features, resolves declared fits through the published ISO 286 zones, and runs worst-case and statistical stack-up analysis.
- AI inference. Model-derived data — extracted feature parameters and rendered views of the part — is sent to Anthropic (Claude) to classify joints and infer assembly intent. See sub-processors below.
- Output. A traceable report, a toleranced NX model and up to four rendered views of the part (PNG) are written to your workspace.
- Deletion. When the run finishes, the compute node deletes the uploaded source file from storage and wipes its own working directory — source file, extracted intermediate data and working renders alike. Your report, toleranced model and the rendered part views kept with them remain until you ask us to delete them.
Data retention
| Data | Retention | Why it exists |
|---|---|---|
| Uploaded .prt source file | Deleted when the analysis completes | Only needed while the analysis runs — re-running afterwards asks for a fresh upload |
| Compute-node working copies | Wiped at end of every run | Transient analysis data on the machine that processes your file |
| Analysis report + toleranced model + rendered part views (PNG) | Until you ask us to delete them | Your deliverable — the reason the tool exists |
| Account (email, display name) | Until you ask us to delete the account | Sign-in, via Firebase Auth |
| Sign-up verification (code hash; salted hash of your network address) | Code hash: until you verify, or at most 52 hours after the code expires; network hash: at most 52 hours | Stop abuse of sign-up emails |
| Free-trial claim signals (email-domain class; salted hashes of your network address and browser fingerprint) | Deleted automatically after 90 days | Limit repeated free-trial claims — hashes only, no raw IP or fingerprint |
| Feedback (first-run survey, KC votes, cancellation reason) | While your account exists | Improve the analysis and decide follow-ups — tied to your account |
| Billing identifiers (Creem customer and subscription IDs, billing period) | While you have a subscription record with us | Match payments to your account — card details stay with Creem |
| Sign-up and payment records (checkout started or failed, Creem notification type and event ID, subscription status changes, failed account emails and file uploads) | Deleted automatically after 180 days | Find where sign-up or payment breaks — tied to your account; no card details, email contents or file contents |
| Email with us — mail to any nxtolerance.com address and every email we send you — plus Team plan inquiries | Until you ask us to delete them — no automatic expiry. An account-deletion request deletes them too | Answering you and following up — kept in Cloudflare D1 (our mail log; Team plan inquiries also in the product database) |
| Operational logs | ≤30 days | Reliability and abuse prevention — metadata only (IDs, sizes, timings, status codes), never model content |
| AI inference payloads | Per Anthropic API terms | Not used for model training; transient abuse-monitoring retention may apply under Anthropic’s policies |
Sub-processors
| Provider | Role | What it can see |
|---|---|---|
| Cloudflare | Hosting, CDN, object storage, edge compute, email receiving and sending, our mail log | Uploaded files at rest; request metadata; email to and from us, and Team plan inquiries |
| Google Firebase | Authentication | Email, display name, auth tokens |
| Anthropic (Claude API) | AI inference — joint/feature classification | Model-derived data: extracted feature parameters and rendered part views. Commercial API terms: inputs are not used for training |
There is no fourth provider hiding anywhere — those three are the whole list of providers that handle your parts and analysis data. Payment is separate: Creem, our merchant of record, handles card details and receipts for paid plans, and we send it nothing from your parts. We keep only its customer and subscription identifiers.
What we deliberately never do
- No training on customer parts. Your uploads are not used to train any model — not ours, not Anthropic’s (their API terms).
- No human review by default. Engineers look at uploaded content only when you explicitly ask for help on a specific analysis.
- No sale or sharing of uploaded files, analysis content, or derived data.
- No advertising or third-party trackers on uploaded data.
Deployment options
- Cloud (default): everything on this page applies.
- On-prem: available for pilot engagements on the Team tier — the pipeline runs inside your infrastructure and nothing, including AI inference payloads, leaves your site. Talk to us.
Honest baseline
SuperNX is not SOC 2 or ISO 27001 certified today. We would rather show you the controls that actually exist — encrypted transport and storage, automatic deletion, a named sub-processor list, least-privilege internal access — than wave a certification we have not earned. Questions and due-diligence checklists: security@nxtolerance.com.
Report a vulnerability
Email security@nxtolerance.com; include reproduction steps and we acknowledge within 72 hours. Machine-readable contact: /.well-known/security.txt (RFC 9116).