Acknowledgements
Open-source software and models · Last updated: 29 August 2026
Blurt stands on other people's work. This page names it, with the licence each piece is used under. The same list is in the app, under Settings → Support → Acknowledgements.
Built with Llama. One of Blurt's optional on-device clean-up models is derived from Meta Llama 3.1 — see the models table below.
Swift packages
Scroll the table sideways to see every column.
| Component | What it does | Licence | Source |
|---|---|---|---|
| WhisperKit — Argmax, Inc. | On-device speech-to-text on the Apple Neural Engine | MIT | github.com/argmaxinc/WhisperKit |
| FluidAudio — Fluid Inference | Runs the Parakeet speech model on the Neural Engine | Apache-2.0 | github.com/FluidInference/FluidAudio |
| SwiftLlama — Shenghai Wang | Swift wrapper around llama.cpp for on-device text clean-up. Blurt uses a patched fork; the changes are in vendor/SwiftLlama | MIT | github.com/ShenghaiWang/SwiftLlama |
| llama.cpp — Georgi Gerganov and contributors | The inference engine behind the local language model (build b5092, as a prebuilt xcframework) | MIT | github.com/ggml-org/llama.cpp |
| Sparkle — Andy Matuschak, Kornel Lesiński, Mayur Pawashe and contributors | Signed automatic updates | MIT | github.com/sparkle-project/Sparkle |
| swift-transformers, swift-jinja — Hugging Face | Tokenizers and chat templates used by WhisperKit | Apache-2.0 | github.com/huggingface/swift-transformers |
| swift-argument-parser, swift-collections, swift-crypto, swift-asn1 — Apple Inc. | Transitive dependencies of the above | Apache-2.0 | github.com/apple |
| yyjson — YaoYuan | JSON parsing, pulled in transitively | MIT | github.com/ibireme/yyjson |
Models
Blurt does not ship model weights inside the app. When you choose an on-device model, the app downloads it from the source below, verifies it against a pinned SHA-256, and stores it in your Application Support folder.
| Model | Role | Licence | Source |
|---|---|---|---|
| Whisper large-v3-turbo and other Whisper variants — OpenAI, converted to Core ML by Argmax | On-device speech-to-text | MIT (weights) · MIT (conversion) | argmaxinc/whisperkit-coreml |
| Parakeet TDT 0.6B v3 — NVIDIA, converted to Core ML by Fluid Inference | On-device speech-to-text with built-in punctuation | CC-BY-4.0 — attribution required, and this is it | nvidia/parakeet-tdt-0.6b-v3 |
| Qwen3-4B-Instruct-2507 — Alibaba Cloud, GGUF quantisation by Unsloth | On-device text clean-up (default) | Apache-2.0 | unsloth/Qwen3-4B-Instruct-2507-GGUF |
| Llama-3.1-Nemotron-Nano-8B-v1 — NVIDIA, built on Meta Llama 3.1; GGUF quantisation by bartowski | On-device text clean-up (optional, larger) | NVIDIA Open Model License, on top of the Llama 3.1 Community License and its Acceptable Use Policy | nvidia/Llama-3.1-Nemotron-Nano-8B-v1 |
Llama 3.1 attribution. The optional Nemotron Nano clean-up model is built with Meta Llama 3.1. Its use is subject to the Llama 3.1 Community License Agreement and the Llama 3.1 Acceptable Use Policy, both of which apply to you when you choose that model. If you would rather not accept them, use the default Qwen3 model instead — it is Apache-2.0 and is what Blurt picks unless you change it.
Cloud models are a different matter: they run on the providers' own machines under the providers' own terms, and are listed on our processor page.
Sounds
Material Design sound resources © Google · CC-BY 4.0. Blurt’s start, stop and event cues are taken from that library. The Glint opening cue is trimmed — the audible attack ended at 0.21 s and the rest was a reverb tail below −45 dB.
The same credit is in the app, under Settings → Support → Acknowledgements.
Typefaces
| Typeface | Designer | Licence |
|---|---|---|
| Hanken Grotesk | Alfredo Marco Pradil | SIL Open Font License 1.1 |
| Newsreader | Production Type | SIL Open Font License 1.1 |
| JetBrains Mono | JetBrains | SIL Open Font License 1.1 |
Backend
The Blurt backend is built on Fastify, Drizzle ORM, postgres.js, jose and ws (all MIT), the official Stripe Node SDK (MIT), PostgreSQL (PostgreSQL Licence) and Caddy (Apache-2.0).
Licence texts
Full licence texts ship with the app and are also available at the source links above. If you would like a copy of any of them by email, write to support-blurt@chbnk.one and we will send it.
If you believe an attribution here is wrong or missing, please tell us — we would rather fix it than argue about it.
Related
See our Terms of Use, Privacy Policy and Impressum.