Every tool on this site runs inside your browser, and almost everything that does the work was written by someone else and released as open source. This page names those projects and their licences. The libraries are bundled into the pages you load from devholster.com. The machine learning models are not: a page that needs one fetches it from huggingface.co when you press its download button, then keeps it in your browser.
Models fetched from Hugging Face
| Model | Used on | Authors | Licence |
|---|---|---|---|
| Whisper | speech to text, subtitle generator | OpenAI; ONNX exports by onnx-community and Xenova on Hugging Face | MIT |
| Kokoro-82M | text to speech | hexgrad; ONNX export by onnx-community | Apache 2.0 |
| BiRefNet_lite | background remover | Peng Zheng et al.; 512-pixel ONNX re-export by Studio Ludens | MIT |
| all-MiniLM-L6-v2 | semantic similarity | sentence-transformers (Nils Reimers et al.); ONNX export by Xenova | Apache 2.0 |
The files come from each model's own repository on huggingface.co, unchanged. Nothing is proxied through this site, and the model repositories are the place to find the weights, the model cards and the licence texts.
eSpeak NG, under the GPL
The text to speech page turns English words into phonemes with eSpeak NG before Kokoro reads them. eSpeak NG is free software under the GNU General Public License, version 3 or later. This site ships it to your browser compiled to WebAssembly, unchanged, as the phonemizer package version 1.2.1 publishes it. The build is loaded as a file of its own, separate from the rest of the page.
Under the GPL you may run, copy, share and change that program. The source code of eSpeak NG is in its repository linked above, the compiled build and its data file are in the phonemizer.js repository, and the licence itself is at gnu.org/licenses/gpl-3.0.html. If you want the source in another form, write to the address in the legal notice.
Libraries in the pages
| Library | What it does here | Licence |
|---|---|---|
| transformers.js | loads and runs the models | Apache 2.0 |
| ONNX Runtime Web | executes the models on WebGPU or WebAssembly | MIT |
| phonemizer.js | eSpeak NG for the browser, see above | Apache 2.0 for the wrapper, GPL 3 for eSpeak NG |
| kokoro-js | the text normalisation and phoneme fix-ups of the text to speech page, ported into our worker and modified there | Apache 2.0, Copyright hexgrad |
| sql-formatter | the SQL formatters | MIT |
| marked | Markdown to HTML | MIT |
| DOMPurify | HTML sanitising | Apache 2.0 or MPL 2.0 |
| jsdiff | the text diff | BSD 3-Clause |
| terser | the JavaScript minifier | BSD 2-Clause |
| csso | the CSS minifier | MIT |
| html-minifier-terser | the HTML minifier | MIT |
| gpt-tokenizer | the token counter | MIT |
| yaml | YAML parsing and formatting | ISC |
| smol-toml | TOML parsing | BSD 3-Clause |
| fast-xml-parser | XML parsing | MIT |
| bcrypt.js | bcrypt hashes | BSD 3-Clause |
| dotted-map | the dotted world map | MIT |
| trystero | the WebRTC connections of the peer to peer chat, video call, file transfer and voice chat, signalled over Nostr relays | MIT |
| uqr | the invite QR code of the peer to peer chat, video call, file transfer and voice chat | MIT |
The typefaces are IBM Plex Mono, Archivo and Wix Madefor Text, all under the SIL Open Font License 1.1 and served from this site. The pages themselves are built with Astro and Tailwind CSS, the social cards and diagrams with Satori, sharp and opentype.js, none of which reach your browser as code.
Each licence above asks that its copyright and permission notice travel with the code. The bundled files carry them where the build keeps comments, and the linked repositories hold the full texts. If you think a project is missing from this page, tell us through the feedback form on any tool page.