Third-party notices

PUSHTYPE is built on open-source software and open speech models.  We're grateful to the people who made them.  This file lists each one, its license and what the license asks of us.  The full license texts are in legal/third-party/, and the release build copies this whole folder into PUSHTYPE.app/Contents/Resources/Legal.

The companies named here have not endorsed PUSHTYPE.  Their names are used only to say where each component comes from.

Software shipped inside PUSHTYPE.app

These are compiled into the app or bundled with it.  Their licenses require us to include the copyright notice and license text with every copy we distribute.

ComponentVersionLicenseCopyrightFull text
FluidAudio0.17.4Apache License 2.0FluidInferencethird-party/FluidAudio/LICENSE.txt
↳ NemoTextProcessing (text-processing-rs), linked by FluidAudio0.3.1Apache License 2.0 (with MIT/Apache sub-components)FluidInference and contributorsthird-party/FluidAudio/NemoTextProcessing-LICENSE.md
↳ fastcluster, compiled into FluidAudio—BSD 2-Clause© 2011 Daniel Müllner; © Google Inc.third-party/FluidAudio/fastcluster-LICENSE.md
↳ VBx, ported in FluidAudio—Apache License 2.0VBx authorsthird-party/FluidAudio/vbx-LICENSE.md
↳ Kokoro text frontends, in FluidAudio—Apache 2.0 / MIT / BSD (see file)Variousthird-party/FluidAudio/JapaneseG2P-LICENSE.md, KokoroAneSpanishFrenchG2P-LICENSE.md
Sparkle2.9.3MIT, plus BSD-style and public-domain components (bsdiff, sais-lite, ed25519, SUSignatureVerifier)Andy Matuschak and the Sparkle contributorsthird-party/Sparkle-LICENSE.txt
LaunchAtLogin-Modern1.1.0MITSindre Sorhusthird-party/LaunchAtLogin-Modern-LICENSE.txt
voice-glow, ported as VoiceGlow.swift—MIT© 2026 Jakub Antalikthird-party/voice-glow-LICENSE.txt
llama.cpp (with ggml), TYPE's text engineb11147MIT© 2023-2026 The ggml authorsthird-party/llama.cpp-LICENSE.txt

We didn't change FluidAudio, Sparkle, LaunchAtLogin, or llama.cpp.  They're used as released (llama.cpp as its official prebuilt framework).  Apache 2.0 §4(b) asks that modified files be marked; since we modified nothing, there is nothing to mark.

Speech models (downloaded on first use, not bundled)

PUSHTYPE doesn't ship model weights.  When you press Download in Settings, PUSHTYPE fetches the model you picked straight from its public Hugging Face repository onto your Mac.  PUSHTYPE never sends your audio anywhere; the models run on your Mac.

We don't redistribute the weights, but we credit them anyway: it's what their authors ask for, and CC BY 4.0 requires it when a work is shared.

In PUSHTYPEModelMade byLicenseNotice
Parakeet Ultramoondream/parakeet-ultra, a post-training of nvidia/parakeet-tdt-0.6b-v3; Core ML conversion FluidInference/parakeet-ultra-coremlMoondream, from NVIDIA's Parakeet TDT 0.6B v3; converted by FluidInferenceCC BY 4.0"Parakeet Ultra" by Moondream, based on "Parakeet TDT 0.6B v3" by NVIDIA, licensed under CC BY 4.0; converted to Core ML by FluidInference.
Parakeet Streamingnvidia/parakeet-unified-en-0.6b; Core ML conversion FluidInference/parakeet-unified-en-0.6b-coremlNVIDIA; converted by FluidInferenceNVIDIA Open Model LicenseLicensed by NVIDIA Corporation under the NVIDIA Open Model License.
Nemotron MultilingualNVIDIA Nemotron 3.5 ASR Streaming Multilingual 0.6B; Core ML conversion FluidInference/Nemotron-3.5-ASR-Streaming-Multilingual-0.6b-CoreMLNVIDIA; converted by FluidInferenceOpenMDW-1.1NVIDIA Nemotron 3.5 ASR Streaming Multilingual, © NVIDIA Corporation, made available under OpenMDW-1.1.
Voice activity detectionSilero VAD; Core ML conversion FluidInference/silero-vad-coremlSilero Team; converted by FluidInferenceMITCopyright (c) 2020-present Silero Team.

Licenses checked on each model's Hugging Face card, 2026-10-14.

Typing models (downloaded on first use, not bundled)

TYPE, the inline autocomplete, runs one of these text models on your Mac through llama.cpp.  As with the speech models, PUSHTYPE downloads the file you pick from Hugging Face; we don't redistribute the weights.  Each is the quantized GGUF build by mradermacher of the original model, and every one is licensed under the Apache License 2.0 (checked on each model card, 2026-10-14).

In PUSHTYPEModelMade byGGUF buildLicense
RecommendedQwen/Qwen3.5-2B-BaseQwen team, Alibaba Cloudmradermacher/Qwen3.5-2B-Base-i1-GGUFApache 2.0
Bestgoogle/gemma-4-E2BGooglemradermacher/gemma-4-E2B-i1-GGUFApache 2.0
Progoogle/gemma-4-E4BGooglemradermacher/gemma-4-E4B-i1-GGUFApache 2.0

All four speech-model licenses (the table before this one) allow commercial use.  What each one requires:

Fonts on pushtype.app

The website serves its own copies of these fonts, so pages load nothing from Google.  All four are under the SIL Open Font License 1.1, which allows this; the fonts are unchanged.

The full license text, with these copyright lines, is served next to the fonts: pushtype.app/assets/fonts/OFL.txt.