Skip to content

Open Source Notices

Last updated: 31 August 2026

Components in the app

Both builds are listed in full, whichever platform you are on: an open-source notice that shows you only your own half is not a notice. Versions are the ones pinned for the current release of each platform.

macOS

These Swift packages are compiled into jhint for macOS.

ComponentVersionCopyrightLicence
WhisperKit — on-device speech recognition0.18.0argmax, inc.MIT
swift-transformers — model download and tokenisation1.1.9Hugging FaceApache-2.0
swift-jinja — template rendering used by the above2.4.2Hugging FaceApache-2.0
swift-crypto — cryptography4.3.1Apple Inc.Apache-2.0
swift-asn1 — ASN.1 encoding, used by swift-crypto1.6.0Apple Inc.Apache-2.0
swift-collections — data structures1.6.0Apple Inc.Apache-2.0
swift-argument-parser — command-line parsing in build tooling1.8.2Apple Inc.Apache-2.0
yyjson — JSON parsing0.12.0YaoYuanMIT
Sparkle — application updates2.9.5Sparkle ProjectMIT

Apache-2.0 components are used unmodified. If we ever ship a patched build of one, the change will be noted here, as that licence requires.

Windows

These packages and libraries travel with jhint for Windows. Three of them are not open source and are listed here because they are redistributed under their own terms, not because their licences are of the same kind.

ComponentVersionCopyrightLicence
Windows App SDK — the application framework and window composition1.6.250228001Microsoft CorporationMicrosoft Software Licence Terms
Microsoft Edge WebView2 — the embedded browser used for web search1.0.3124.44Microsoft CorporationBSD-3-Clause
Windows Community Toolkit — renders the assistant answer as Markdown7.1.2.NET Foundation and ContributorsMIT
Bouncy Castle — checks the signature of the licence token and of update packages2.4.0The Legion of the Bouncy Castle Inc.MIT
Open XML SDK — reads .docx files you attach to a session3.5.1Microsoft CorporationMIT
Velopack — installs and updates the application0.0.1298 (library), packaged with vpk 1.2.0Velopack Ltd and contributorsMIT
whisper.cpp — recognises speech on this computercommit c122757fddf358397bb7f33b6ac3aab24a5bca04The ggml authorsMIT
llama.cpp — runs the built-in answer model on this computercommit 6d05498314db1b57f81c271080018aa2d0b89be9The ggml authorsMIT
ColorCode — colours code blocks inside the answer2.0.13.NET Foundation and ContributorsMIT
NuGet.Versioning — compares version numbers when an update arrives6.14.0Microsoft CorporationApache-2.0
Windows SDK .NET projections — let the application call Windows itself10.0.22621.57Microsoft CorporationMicrosoft Windows SDK Licence Terms

The .NET runtime is not in that table on purpose. jhint for Windows is built framework-dependent: the runtime is not shipped inside the app, and the installer fetches it from Microsoft if the computer doesn’t already have it. It is Microsoft’s software under Microsoft Software Licence Terms, and we redistribute none of it.

The app ships the same list with the full text of every licence and upstream notice: Settings → General → Diagnostics → Open Source Licences. That file is assembled at build time from the packages of the very build being released, so it cannot drift from what you actually run.

Three of the entries above — ColorCode, NuGet.Versioning and the Windows SDK projections — were identified after 1.8.3 was published and reach the in-app file with the next release. None of them carries a copyleft term.

The bundled inference engine

The built-in model runs on llama.cpp, built from source at a pinned commit and shipped inside the app — on macOS as the project’s llama-server binary, on Windows as the same engine linked into the app. These are the components actually compiled into it — not the project’s full component list, but the ones our build contains:

ComponentCopyrightLicence
llama.cpp and ggml — the engine itselfThe ggml authorsMIT
nlohmann/json — JSON parsingNiels LohmannMIT
cpp-httplib — HTTP serverYuji HiroseMIT
xxHash — non-cryptographic hashingYann ColletBSD-2-Clause
SHA-256Igor PavlovPublic domain
SHA-1Steve ReidPublic domain
rotate-bitsWilliam CasarinMIT
stb_image — image decodingSean BarrettMIT or public domain, at your choice
miniaudio — audio decodingDavid ReidPublic domain or MIT-0, at your choice
subprocess.hNeil HenningUnlicense (public domain)
llamafile sgemm — matrix multiplication kernelsMozilla FoundationMIT
YaRN context-extension kernelsJeffrey Quesnelle and Bowen PengMIT

The build is deliberately narrow: no OpenSSL, no libcurl and no web interface — the server we ship answers API requests from the app and nothing else. Where a component offers a choice of licences, the app carries both texts and we claim no choice between them.

Language models for the built-in engine

Model weights are not part of the app: the app downloads them from Hugging Face when you choose a model, from a pinned repository revision, and verifies them against a checksum. We do not train, own or republish them.

ModelRepositoryLicence
Qwen3 4B Instruct 2507unsloth/Qwen3-4B-Instruct-2507-GGUFApache-2.0
Qwen3 VL 4B Instructunsloth/Qwen3-VL-4B-Instruct-GGUFApache-2.0
Qwen3 8Bunsloth/Qwen3-8B-GGUFApache-2.0

jhint offers only models published under Apache-2.0 or MIT. After a download, the app writes the full licence text and a PROVENANCE.txt — repository, revision, source address, checksum and size — into the model’s folder on your computer, so what you have on disk says for itself where it came from and under what terms.

The full licence texts ship with the app. They are in the application bundle and you can read them in Settings → General → Open Source Licences — the Swift packages above, the engine’s components, and the upstream NOTICE files that Apache-2.0 requires (swift-crypto and swift-asn1 carry one).

Speech-recognition models

Recognition uses Whisper models by OpenAI, published under the MIT licence. On macOS they are distributed in Core ML form through the whisperkit-coreml repository maintained by argmax; on Windows they run through whisper.cpp in the ggml format published by that project. Either way the app downloads the model chosen for your machine on first use and keeps it in a folder on your computer.

On this website

Two typefaces are used, both under the SIL Open Font License 1.1 and served from this site, not from a font CDN:

TypefaceUsed forCopyrightLicence
OnestBody text and headingsCopyright 2021 The Onest Project AuthorsOFL-1.1
JetBrains MonoTechnical microcopyCopyright 2020 The JetBrains Mono Project AuthorsOFL-1.1

Both files served here are subsets of the originals — Latin characters only, and a single weight of JetBrains Mono — made so pages load faster. Neither typeface declares a Reserved Font Name in the licence it ships with, so the subsets keep the original family names. The full licence text of each is linked in the table above and served from this site alongside the font.

The site itself is built with Astro (MIT) and Tailwind CSS (MIT). These are build tools: their code is not redistributed as a library, only the generated HTML and CSS.

Paddle’s checkout script is loaded on the pricing and checkout pages. It is proprietary software belonging to Paddle, not an open-source component.

Licence texts

The MIT, Apache-2.0 and OFL-1.1 licences all require that copyright notices and licence text travel with the software. For the app, they are inside it — Settings → General → Open Source Licences on macOS, Settings → General → Diagnostics → Open Source Licences on Windows. For the two typefaces on this website, they are served from this site next to the font files and linked in the table above.

If you need the licence texts in another form, or spot something missing from this list, write to support@jhint.ai and we will fix it.

Questions about this document? Write to support@jhint.ai .