Acknowledgments
SideWave runs on models and code that other people published. This page lists each one, what SideWave uses it for, and its license. The app has the same list in Settings > About > Acknowledgments, with every license and notice file in full.
Speech models
FluidInference converted these to Core ML so they run on the Neural Engine. The conversions don't retrain or change the models.
NVIDIA Parakeet TDT v3 (0.6B)
Confirms what you said, one segment at a time. By NVIDIA, licensed under CC-BY-4.0. Model card.
NVIDIA Parakeet Realtime EOU 120M
Writes the gray live draft and notices when you stop talking. By NVIDIA, licensed under the NVIDIA Open Model License. Model card.
Parakeet Ultra
An optional recognition model, post-trained by Moondream from NVIDIA Parakeet. Licensed under CC-BY-4.0. Model card.
NVIDIA Parakeet TDT v2 (0.6B)
An optional English-only recognition model. By NVIDIA, licensed under CC-BY-4.0. Model card.
NVIDIA Parakeet TDT-CTC 110M
Checks finished segments against your glossary terms. By NVIDIA, licensed under CC-BY-4.0. Model card.
NVIDIA Parakeet Unified EN 0.6B
The live transcript during a meeting, on Macs with 16 GB of memory or more. Downloaded only when you ask for it. By NVIDIA, licensed under the NVIDIA Open Model License. Model card.
NVIDIA Nemotron 3 Diarization
Labels up to eight speakers in a meeting. By NVIDIA, licensed under OpenMDW-1.1. Model card and license.
Silero VAD v6
Finds where speech starts and stops. By the Silero team, licensed under MIT. Source.
Language models
Gemma 4 (E2B, E4B, and 12B)
Refine, document edits, vocabulary suggestions, and meeting notes. By Google DeepMind, licensed under Apache 2.0. SideWave downloads the 4-bit MLX conversions from mlx-community when you pick one.
Apple Foundation Models
Refine on Macs with Apple Intelligence turned on. The model comes with macOS 26 and later, and SideWave calls it through the Foundation Models framework. SideWave downloads nothing for it.
Code
| Project | Used for | License |
|---|---|---|
| FluidAudio by FluidInference | Running the speech models, voice detection, and speaker labels | Apache 2.0 |
| NeMo Text Processing by NVIDIA, through FluidAudio | Turning spoken numbers, dates, and amounts into digits | Apache 2.0 |
| MLX Swift and MLX Swift LM | Running Gemma on the GPU | MIT |
| Swift Transformers and Swift Jinja | Tokenizers and chat templates | Apache 2.0 |
| Swift Hugging Face | Model downloads | Apache 2.0 |
| Swift Markdown and cmark-gfm | Documents and meeting notes | Apache 2.0 with Runtime Library Exception; BSD-2-Clause and MIT |
| Sparkle | Updates | MIT |
Those projects bring in smaller packages: Apple's Swift Collections, Swift Numerics, Swift Atomics, SwiftNIO, and Swift Crypto (all Apache 2.0), plus yyjson and EventSource (both MIT). MLX Swift includes MLX, MLX C, {fmt}, and JSON for Modern C++ (all MIT) and Apple's metal-cpp (Apache 2.0). FluidAudio includes fastcluster (BSD), VBx (Apache 2.0), and its Kokoro text front ends, each with its own notice. Swift Syntax is used only while building, and Swift System and Swift ASN.1 are resolved but not linked.
Apple frameworks
SideWave is written in Swift with SwiftUI. It uses Core ML for the speech models, AVFoundation for audio, and AppKit for the shortcut, the floating preview, and pasting.
Something missing?
If an attribution here is wrong or missing, open an issue and I'll fix it.