Thank you

Nasar Flow stands on other people's work.

Every model and library below is open, freely shared, and doing real work inside the app right now. None of this exists without the people who built and published it. Thank you.

Inference engine

whisper.cpp

The on-device engine Nasar Flow runs on. A full C/C++ port of Whisper that makes real-time, offline transcription possible on an iPhone, built and maintained by Georgi Gerganov and the ggml-org community.

github.com/ggml-org/whisper.cpp →
Original research & model

Whisper

The speech recognition model and research this entire app is built on, released openly by OpenAI. Every model Nasar Flow uses traces back to this work.

github.com/openai/whisper →
Singlish model

whisper-small-singlish-122k

The fine-tuned model behind Nasar Flow's default Singlish and English transcription, trained on 122k samples of real Singlish speech and shared publicly by jensenlwt.

huggingface.co/jensenlwt →
Arabic model

whisper-small-arabic-dialectal

The colloquial (Amiyah) Arabic model powering Nasar Flow's Arabic support, shared by oddadmix. It's still early and improving, and that's exactly why the beta exists.

huggingface.co/oddadmix →
Build tooling

XcodeGen

Generates the Xcode project from a plain text spec on every CI run, which is exactly what makes it possible to build and ship this whole app from GitHub Actions without ever opening Xcode on a Mac.

github.com/yonaskolb/XcodeGen →

Nasar Flow's own source is public too, under the MIT license. See it on GitHub →