Third-party notices.

Last updated: April 30, 2026

Local model roles

Liptyper uses two local model roles: a speech recognition model for transcription and a text cleanup model for spacing, punctuation, capitalization, and joining dictated snippets. These models run on the user's Mac after download.

Speech recognition

The default speech model is mlx-community/whisper-large-v3-turbo, an MLX-format community model derived from OpenAI's Whisper large-v3-turbo family. OpenAI's Whisper project is distributed under the MIT License, and the OpenAI Hugging Face model card for whisper-large-v3-turbo is tagged MIT.

Text cleanup

The default cleanup model is mlx-community/Qwen3.5-0.8B-MLX-8bit, a quantized MLX version of Qwen/Qwen3.5-0.8B. Its model card lists Apache 2.0.

Libraries and fonts

Liptyper also uses open-source Python and macOS libraries, including MLX, mlx-whisper, mlx-lm, PyObjC, sounddevice, rumps, Hugging Face Hub tooling, Transformers, Tokenizers, NumPy, and tomli-w. Outfit is used under the SIL Open Font License.

No endorsement

References to Apple, OpenAI, Qwen, MLX, Hugging Face, and other projects are for attribution and compatibility context only. Liptyper is not affiliated with, sponsored by, or endorsed by those projects or companies.

More detail

A fuller third-party notices file is included with release materials. Questions: [email protected]