What others have said and built
Open-source claims are cheap; what other people did with the code is not. These are the third-party signals, each with a link.
- Hacker News: the engineering write-up “Run CLIP on iPhone to Search Photos” reached the front page in February 2023 with 397 points and 134 comments (thread). The developer submitted it; the points and comments are readers’.
- GitHub: 2,994 stars, 445 forks and six contributors, checked October 30, 2026 (repository).
- Android: PicQuery by greyovo, itself MIT-licensed, names Queryable in its README as “the inspiration for this project” (PicQuery). The PyTorch Korea write-up below also names it as the Android counterpart.
- Press: the German Apple site iphone-ticker.de reviewed it on February 8, 2023 as an impressive local photo search for the iPhone (review).
- Community: the PyTorch Korea forum published a write-up of the app and its MobileCLIP model on January 25, 2025 (PyTorch Korea).
Public repository vs the App Store app
Be clear about what the public code is. The repository carries the photo-search core — indexing, the MobileCLIP and earlier CLIP encoders, embedding storage and GPU ranking. The App Store app is built from a private repository that has moved ahead of it.
Features added in the 2.1.x releases — translated queries, search by photo, time filters, batch actions and the similar-photo cleanup — are not in the public code. On October 30, 2026 the last push to the public repository was dated March 29, 2026, and it has no tagged releases. Reading the public code tells you how the search core works and that its design is on-device; it does not let you audit every line of the current binary.
What the repository lets you verify
Privacy claims are easier to judge when the code is public. The source shows where photos are requested, how images and text become embeddings, where those embeddings are stored and how a query is ranked, so you can check the search core yourself instead of trusting a marketing line. The design is explained on how Queryable works.
It also exposes practical engineering trade-offs: initial indexing can be demanding, iCloud assets may fail to fetch, large libraries need careful memory handling, and similarity models sometimes return imperfect matches.
- Swift and SwiftUI application code
- Core ML model wrappers for MobileCLIP-S2
- Incremental embedding storage
- Metal/GPU similarity ranking
- Photo-library actions
What the MIT license permits
The MIT license is permissive: developers can use, modify, distribute and study the covered source subject to its short notice and warranty terms. Model files and Apple frameworks can have their own terms, so the repository license should not be treated as a blanket license for every external dependency.
Open source also does not mean the App Store binary is free of charge. Queryable is sold as a one-time paid app ($4.99 in the US); purchasing supports the maintained release while the public code remains available for inspection and learning.
How to evaluate or contribute
Start with the README and license, then trace the photo-search view model into the image encoder, text encoder, embedding store and GPU search implementation. When reporting a bug, include the device model, iOS version, approximate library size, iCloud setting and a reproducible sequence without attaching private photos.
Contributions should preserve the app’s privacy model and be measured on real-device constraints. A feature that works on a tiny simulator library may behave very differently with tens of thousands of iCloud-backed assets.
Other open-source photo search
Queryable is one open-source option among several, and the right one depends on where your photos live. Ente Photos is open source under AGPL, encrypts backups end to end, runs MobileCLIP search on the device and now works without an account (Queryable vs Ente Photos). Immich and PhotoPrism are open-source, self-hosted photo servers with smart search (Queryable vs Immich). Queryable is the one that needs no server and no new gallery: it searches the Apple Photos library already on the phone. The full list is on alternatives.
What open source does—and does not—prove
Public code makes review possible; it does not guarantee that every line has been audited or that every build is free of defects. Users should still review App Store privacy information, requested permissions and network behavior. Queryable’s App Store privacy label reads “Data Not Collected”.
For Queryable, open source is strongest as corroborating evidence: the published architecture is consistent with an on-device semantic search design, and the core implementation can be challenged or improved in public.
Common questions
Where is Queryable’s source code?
The official public repository is github.com/mazzzystar/Queryable, under the MIT license.
Is the App Store app built from the public code?
From a private repository that is ahead of the public one. The public code carries the search core; features added in 2.1.x, such as translated queries, search by photo, time filters and similar-photo cleanup, are not in it.
Is Queryable the only open-source photo search app for iPhone?
No. Ente Photos is open source and offers on-device MobileCLIP search, and Immich and PhotoPrism are open-source self-hosted servers. Queryable’s search core is open source; it needs no server and searches the Apple Photos library you already have, though the App Store build adds features that are not in the public code.
Can I build my own version?
The covered source is MIT-licensed, but you must follow Apple developer requirements and the separate terms of any models or third-party assets.
Why pay if the core is open source?
The App Store app is the maintained build, with features the public code does not have — translated queries, search by photo, time filters and similar-photo cleanup — and buying it pays for that work.
Primary sources and provenance
Where the facts on this page come from. Anything from the 2022 prototype is marked as history, not as how the app works today.