EditToolbelt is built on open-source software. This list is generated from our license register and grows as tools go live. No GPL code ships to your browser; LGPL parts are separate files, with a link to their source.
Platform and app
| Project | License | Used for |
|---|
| Next.js, React | MIT | web app |
| Tailwind CSS | MIT | styling |
| Zod | MIT | validation |
| Drizzle ORM | Apache-2.0 | database schema, queries and migrations (packages/db) |
| node-postgres | MIT | Postgres driver for the web server and scripts (packages/db) |
| Better Auth | MIT | accounts: magic-link and Google sign-in, sessions, TOTP for admins (server only; its telemetry is switched off) |
| Nodemailer | MIT-0 | sends sign-in emails over SMTP (Mailpit on the local stack) |
| aws4fetch | MIT | signs S3 requests and presigns upload and download URLs for object storage (server only; no dependencies) |
| pino | MIT | logging |
| structlog | MIT OR Apache-2.0 | Python logging |
| Modal client (`modal`) | Apache-2.0 | the worker calls the GPU functions on Modal; CI deploys the Modal app (ServerlessGpu, docs/01 → GPU backend) |
| Lucide | ISC | icons |
| Onest (variable) + IBM Plex Mono | OFL-1.1 | UI and numeric fonts, self-hosted |
| Fontsource packages for Onest and IBM Plex Mono | OFL-1.1 | source of the self-hosted font files (copied into the build, subset by unicode-range); static Onest weights for build-time OG images |
| Fontsource packages for Montserrat, Oswald and Noto Serif | OFL-1.1 | font files for text on images (P10), 400 and 700, Latin, Latin Extended, Cyrillic and Cyrillic Extended; copied into the build and loaded only when picked. Chosen because each has Uzbek Latin oʻ gʻ (U+02BB) and Uzbek Cyrillic қ ғ ҳ ў |
| Pillow | MIT-CMU | server image probe/ops; reads, resizes and writes images and frames in the GPU images (Upscale Image, Object Eraser, Video Background Remover) |
Browser processing
| Project | License | Used for |
|---|
| Mediabunny | MPL-2.0 | video and audio tools (V01, V02, V04, V06): demux and mux MP4, MOV, WebM, MKV, MP3, WAV, OGG, FLAC, ADTS; drives the browser’s WebCodecs; no codecs of its own |
| jSquash (avif, webp, jpeg, png, oxipng, resize) | Apache-2.0 | image codecs (P05, P06): WASM builds of MozJPEG/libjpeg-turbo (IJG + BSD-3-Clause), libwebp (BSD-3-Clause), libavif + libaom (BSD-2-Clause, AOM royalty-free patent licence), the png crate and OxiPNG (MIT); single-threaded builds only |
| onnxruntime-web | MIT | browser ML |
| qrcode (npm) | MIT | QR generation (U01): the browser entry only, which builds the module matrix; we render SVG and PNG ourselves. Its CLI dependencies (yargs, pngjs) are never shipped |
| fflate | MIT | ZIP of batch outputs ("Download all") and of Split Image's tiles, in the image worker; deflating Images to PDF's non-JPEG images |
| hash-wasm | MIT | File Checksum's MD5, SHA-1 and SHA-256, streamed in WebAssembly so multi-GB files hash in the browser (WebCrypto has no MD5 and no streaming) |
| @mediabunny/mp3-encoder | MPL-2.0 | MP3 encoding for Extract Audio (WebCodecs has no MP3 encoder): a WASM build of the LAME encoder It includes LGPL code, loaded as a separate file you can replace; get its source code and how our copy is built. |
| @mediabunny/flac-encoder | MPL-2.0 | FLAC encoding for Extract Audio (no browser encodes FLAC): a WASM build of libFLAC (BSD-3-Clause) |
Server tools
| Project | License | Used for |
|---|
| Noto fonts (fonts-noto-core) | OFL-1.1 | the fonts Burn Subtitles draws text with (Noto Sans, Serif and Sans Mono), Debian's package in the worker image |
| Pydantic, pydantic-settings | MIT | worker env and job payload validation (Python side of the Zod rule) |
| boto3 | Apache-2.0 | S3/R2 client for the worker |
GPU images (Modal, server only)
| Project | License | Used for |
|---|
| PyTorch (`torch`, `torchvision`) | BSD-3-Clause | runs the models on the GPU, inside the Modal images only |
| ONNX Runtime GPU (`onnxruntime-gpu`) | MIT | runs MI-GAN (P17) and BiRefNet (V21) on the GPU, inside the Modal images only (gpu/requirements-onnx.txt, hashed); its own dependencies are permissive: flatbuffers (Apache-2.0), protobuf (BSD-3-Clause), packaging (Apache-2.0 OR BSD-2-Clause) |
| openai-whisper | MIT | Whisper's reference implementation (A12, V17), inside the Modal image only |
| numba | BSD-2-Clause | Whisper's word timing, inside the Modal image only |
| spandrel | MIT | loads the Real-ESRGAN networks from their weights (P08), inside the Modal image only |
| NumPy | BSD-3-Clause | arrays, inside the Modal images only |