Privacy Policy
PicSift processes your photos entirely on your device. We don't collect your photos, scores, or personal information. We don't run analytics. We don't have servers that see your data.
1. Data We Access
Photos and Metadata
PicSift accesses photos you select through the macOS/iOS file picker or the Photos library (with your permission). For each photo, PicSift reads:
- Image pixel data (for scoring and display)
- EXIF metadata (camera model, focal length, ISO, aperture, shutter speed, capture date)
- Location data embedded in photo EXIF (latitude/longitude), if present
- File path and filename
This data produces quality scores, detects scenes, detects and groups faces, finds duplicates, and powers search. All of that analysis runs on your device. Your photos are never sent to any server. Section 6 describes the models and frameworks that do the work.
One piece of metadata leaves your device. If a photo has location data, PicSift sends its GPS coordinates to Apple's map service. This shows a place name such as “Seattle, WA”. It sends only the coordinates. Section 4 lists this connection.
Face Data
PicSift uses on-device machine learning to detect faces. This produces face bounding boxes, face embeddings (numbers used to group faces) and quality measures. Face data is stored on the device and is never sent off it. To delete all face data on a Mac, choose Settings → Advanced → Reset to First Run. On iPad, delete the app.
Curation Decisions
Your Keep, Reject, and Delete Tray decisions, star ratings, tags, captions, and Smart Album settings are stored on the device.
2. Data We Store
PicSift keeps a SQLite database on the device. It holds photo file paths, quality scores, scene categories, face embeddings, search embeddings, curation decisions, and Smart Album settings. PicSift also keeps previews and thumbnails of your photos, and on iPad, copies of the photos you import. All of this exists only on the device. You can delete it at any time.
3. No Cloud Sync
PicSift does not sync anything between devices. Your scores, decisions, star ratings, tags and Smart Albums live only in the local database of the device that made them. PicSift on a Mac and PicSift on an iPad each keep their own. Nothing is written to iCloud. There is no sync setting to turn on.
4. Network Connections
| Connection | Purpose | Frequency |
|---|---|---|
| GitHub | Fetch the current list of available models and scoring presets | On app launch |
| jsDelivr (a public mirror of GitHub) | The same lists, used only if GitHub cannot be reached | Only as a fallback |
| GitHub Releases | Download optional ML models | When you choose to download one |
| Apple Maps (MapKit) | Turn a photo’s GPS coordinates into a place name, and draw the Map view | When photos with location data are scored or mapped |
| Apple StoreKit | Verify purchases | On app launch |
None of these requests carries a photo, a score or an account. The Apple Maps request carries a photo’s coordinates and is handled under Apple’s privacy policy. PicSift does not contact any Syrinx Ventures servers.
5. Data We Do Not Collect
- No analytics or usage telemetry
- No third-party analytics services
- No crash reports are collected. Crash and performance reports stay on the device unless you share one.
- No device identifiers, IP addresses, or location data
- No user accounts or profiles
- No advertising
- No selling, sharing, or transferring data to third parties
6. How Photos Are Analyzed
PicSift analyzes photos with three kinds of tools. All of them run on your device, and no photo or result is sent to a server to be analyzed.
- Open research models. These are models published by Google, Apple and university research labs, converted to run through Core ML. Core ML is Apple’s framework for running machine-learning models on the Mac’s or iPad’s own chip. It is the engine, and these are the models it runs. Three ship inside the app. ARNIQA (a university lab) and MUSIQ (Google) score Overall Appeal. ARNIQA runs on Macs and M-series iPads. MUSIQ runs on other iPads. AdaFace groups faces into people. The optional downloads are SigLIP 2 (Google), which runs Visual Search, and NIMA (Google), another engine for Overall Appeal. Apple’s MobileCLIP runs Visual Search on a device that has it instead of SigLIP 2. The in-app Model Library names each model with its creator and license. Each entry links to that model's paper or project.
- Apple’s Vision framework. It is built into macOS and iPadOS. It detects faces and judges their quality. It finds the main subject of a photo and classifies scenes. It makes the image fingerprints behind “More Like This” and supplies Apple’s own aesthetic score.
- Plain image measurement. This uses no machine learning. It measures sharpness from how crisp the edges in a photo are. It measures exposure from its brightness histogram. It runs duplicate detection from a compact fingerprint of each image.
Optional models are downloaded from GitHub over a standard HTTPS request, which sends nothing about your photos.
7. Children's Privacy
PicSift does not knowingly collect personal information from children under 13. The app does not require an account and processes everything locally.
8. Data Retention and Deletion
All data is stored on the device. To remove all data on a Mac, use Settings → Advanced → Reset to First Run. On an iPad, delete the app. Syrinx Ventures keeps no data on any server.
9. Your Rights
PicSift collects no personal data and sends none to our servers, so you keep full control. On a Mac, the data is a SQLite file you can inspect, back up or delete. On iPad, deleting the app deletes the data.
10. Changes to This Policy
We may update this policy from time to time. Material changes will be noted in PicSift's release notes.
Privacy questions
- privacy@picsift.com PicSift is published by Syrinx Ventures, LLC.