Last updated: 2026-05-07
Privacy Policy
The short version
Your inspections never leave your iPhone. Every AI feature — voice transcription, summarization, severity classification, recommendations, report generation — runs locally on your device. We do not have servers that store, process, or even see your data. No cloud, no analytics, no third parties, no exceptions.
What we never collect
- Voice recordings, audio, or transcripts
- Photos, scanned documents, or attached files
- Inspection content, findings, severities, or recommendations
- Property addresses, client names, or business information
- App usage analytics, behavioral data, or telemetry
- Crash reports that include any of your data
- Advertising identifiers (we don't run ads)
- Device fingerprinting or any tracking signals
We don't have a database. We don't have a backend. There is nothing on our end to breach because there is nothing on our end at all.
What stays on your iPhone
- Voice recordings, transcripts, photos, scanned documents, and attached files
- Inspection reports, sections, findings, and severity classifications
- Your business profile (name, logo, contact information) used to brand PDFs
- AI summaries and structured findings — generated locally by an on-device language model
These are stored in the iOS app's sandbox and (optionally) backed up to your iCloud Drive under your Apple ID. We never see them.
Where the AI runs
The AI model used for every intelligent feature in OnSiteNotes runs entirely on your iPhone's Neural Engine. The model is Gemma 3, an open-weight model released by Google. It generates structured findings, polished descriptions, severity classifications, and recommendations directly from your voice notes and inputs — without contacting any server, ever.
Voice-to-text transcription uses Apple's on-device SFSpeechRecognizer
with the requiresOnDeviceRecognition
flag set, so even Apple's servers never receive your audio.
The one network call we make
On first launch only, OnSiteNotes downloads the AI model file from
Hugging Face's public model registry (huggingface.co).
This is identical to downloading any file from any public CDN: Hugging Face sees your
IP address and a generic User-Agent header — nothing about you, your reports, or your
usage. We send no data with this request beyond what every web browser sends when
you visit a URL.
After that one-time download, the app makes zero network calls. You can put your iPhone in airplane mode and OnSiteNotes works exactly the same.
Total data ownership
Your inspection data is yours. Period. You don't need to delete an account, export anything, or wait for a "data deletion request" to be processed. To remove all your OnSiteNotes data from existence, delete the app from your iPhone. Done.
Third-party software credit
- Gemma 3 — Google's open-weight model, used under the Gemma Terms of Use.
- Apple Speech framework — on-device transcription.
- MLX — Apple's open-source machine-learning framework that runs the model on the Neural Engine.
- Hugging Face Hub — public model registry (used only for the one-time download).
App Store payments
Subscriptions are processed by Apple. We never see your card details. Apple shares only an opaque receipt with us so we can validate your subscription status. The receipt does not contain any personal information.
Children
OnSiteNotes is a B2B tool for trade professionals. We don't market to children and do not knowingly collect data from anyone under 13.
Changes
If we ever change how data is handled in the app, we will publish the change on this page, bump the "Last updated" date, and surface a clear in-app notice. We will never retroactively transmit existing user data anywhere.
Contact
Questions: [email protected]