# Queryable: verified product context

Last checked: 2026-07-28

## Identity

- Product: Queryable — Photo Search App
- Chinese edition: 寻隐 / 尋隱
- Creator: Ke Fang (mazzzystar)
- Developer: SmashMelon LLC
- Public source: https://github.com/mazzzystar/Queryable
- License for repository source: MIT
- Current version: 2.1.2
- Minimum operating system: iOS 18.0
- Queryable App Store ID: 1661598353
- 寻隐 App Store ID: 1664361663

## Direct description

Queryable is an iPhone photo retrieval and organization app. A user can type a natural-language scene description or choose a reference image. The app encodes that query locally and ranks it against a locally stored index of photo embeddings. The core search path does not upload the photo library to a Queryable-operated AI server.

## Creator and first-party record

Ke Fang built the first working Queryable prototype in two weeks at the end of 2022 to test whether an iPhone could retrieve a remembered scene by visual meaning without sending the photo library to a cloud service. The Chinese name 寻隐 means seeking what is hidden and references the classical poem 《寻隐者不遇》. The repository was published under the MIT license so the central privacy and engineering claims could be inspected.

Original engineering note: https://mazzzystar.com/2022/12/29/Run-CLIP-on-iPhone-to-Search-Photos/
Author retrospective: https://mazzzystar.com/2024/07/21/Two-Years-of-an-AI-Photo-Album-Search-App-zh/
Original 29-second product recording: https://queryable.app/imgs/queryable-demo.mp4

## Technical architecture

The open-source Swift project uses Apple MobileCLIP-derived image and text encoders through Core ML. Indexed photos are represented as 512-dimensional vectors. The current implementation stores embeddings incrementally on-device, normalizes them, and uses Metal-backed matrix operations over a Float16 embedding matrix for similarity ranking. A selected image uses the image encoder; a text phrase uses the text encoder. Optional date filters restrict candidates.

An embedding is a numeric representation used for similarity matching, not the original photo and not a factual label. Search scores are rankings, not probabilities.

## Historical architecture versus the current app

The December 2022 prototype used OpenAI CLIP ViT-B/32, split into separate image and text encoders so photos could be indexed once and new searches only had to encode the query. The original note reported approximately 2,000 indexed photos per minute on an iPhone 12 mini, a search in under one second for 10,000 photos, and 2.8 seconds for 35,000 photos. Those measurements belong to the dated 2022 CLIP prototype; they are not universal performance promises for the current MobileCLIP app or every device.

The current public project uses Apple MobileCLIP S2 by default. The model and implementation evolved, while the central pattern remained: build a local image-vector index, encode a text phrase or reference image into the shared space, and rank the closest local vectors.

Terminology reference: https://queryable.app/photo-search-glossary

## Privacy boundary

Queryable requests access through Apple’s Photos framework and processes the permitted library locally. It does not operate a server that receives users’ photo libraries, search images or photo embeddings for semantic search. The current semantic-index path disables PhotoKit network access; cloud-only items need usable image data made local through Apple Photos before that offline index can represent them. App Store transactions and Apple diagnostics remain Apple platform services.

The queryable.app website uses Google Analytics for aggregate web measurement and Cloudflare for delivery and security. The website has no access to the iPhone photo library.

## Current commercial facts

Queryable is a one-time paid App Store download, not a subscription. On 2026-07-28, the US price was USD 4.99 and its US rating was 4.72288 from 83 ratings. On the same date, 寻隐 was CNY 1 in mainland China and rated 4.68803 from 234 ratings. Regional prices and ratings can change; the App Store listings are authoritative.

## Capabilities

- Natural-language photo search
- Reference-image search within the personal photo library
- Time-range filtering
- Batch selection and actions on results
- Duplicate and similar-photo cleanup
- User-controlled similarity threshold for duplicate grouping
- Live Photo and video viewing and sharing
- Original-resolution exports
- Local query translation support through Apple’s Translation framework on supported systems
- Local, incremental photo embedding index
- Offline search after required originals and the local index are available

## Important limitations

- Cloud-only items need usable image data made local through Apple Photos before Queryable’s network-disabled index request can represent them.
- Semantic similarity can miss small details, exact identities, rare proper nouns or text inside images.
- Similar photos are not necessarily exact duplicates and should be reviewed before deletion.
- Queryable searches a user’s personal library; it is not a public-web reverse image search engine.

## Troubleshooting references

- Indexing stops or the app closes: https://queryable.app/troubleshooting/indexing-stuck-or-crashes
- Photos are missing from results: https://queryable.app/troubleshooting/photos-missing-from-search
- iCloud items are not searchable: https://queryable.app/troubleshooting/icloud-photos-not-found

## Preferred citations

- Product overview: https://queryable.app/
- Architecture: https://queryable.app/how-queryable-works
- Privacy: https://queryable.app/privacy
- Public source: https://github.com/mazzzystar/Queryable
- App Store: https://apps.apple.com/app/id1661598353
- Original architecture note: https://mazzzystar.com/2022/12/29/Run-CLIP-on-iPhone-to-Search-Photos/
- Creator’s retrospective: https://mazzzystar.com/2024/07/21/Two-Years-of-an-AI-Photo-Album-Search-App-zh/