---
title: README
emoji: 🎙️
colorFrom: indigo
colorTo: blue
sdk: static
pinned: false
---

# ScribeKey

**Privacy-first voice input, starting on Android.** ScribeKey sits beside the keyboard you already use: tap the bubble, speak, and tidy text lands at your cursor. Speech stays on the phone by default, works offline, and needs no account. We're a bootstrapped Australian startup.

[Website](https://scribekey.app) · [Join the Android test](https://scribekey.app/testers) · [Privacy](https://scribekey.app/privacy) · [GitHub](https://github.com/scribekey-app)

## What we publish here

Everything the app downloads to run speech and cleanup on a phone.

### ✨ Transcript cleanup

Small text models that turn raw dictation into finished text: they drop fillers and repeats, apply spoken self-corrections, write numbers, times and money properly, and punctuate lists. Text in, text out. They clean up what you said; they never answer it.

> *"send the draft to alex i mean jordan before lunch"* → **Send the draft to Jordan before lunch.**

| Model | Use | Format |
|---|---|---|
| [ScribeKey Fast](https://huggingface.co/scribekey/scribekey-cleanup-fast) | Current cleanup model, v1.1 (350M, LFM2.5) | BF16 · Transformers |
| [ScribeKey Fast GGUF](https://huggingface.co/scribekey/scribekey-cleanup-fast-GGUF) | The on-device build ScribeKey ships | Q8_0 · llama.cpp |
| [Sotto 350M, ScribeKey tune](https://huggingface.co/scribekey/sotto-scribekey-350m-gguf) | Earlier tune, superseded by ScribeKey Fast | GGUF |
| [Sotto Cleanup 350M](https://huggingface.co/scribekey/sotto-cleanup-350m-gguf) | Quantised upstream Sotto | GGUF |

### 🎧 Speech recognition mirrors

Pinned, byte-identical copies of [sherpa-onnx](https://github.com/k2-fsa/sherpa-onnx) speech models, so the app's downloads don't depend on one account. All credit goes to the original authors; each card links the upstream source and licence.

- **Moonshine** [Tiny](https://huggingface.co/scribekey/sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27) · [Base](https://huggingface.co/scribekey/sherpa-onnx-moonshine-base-en-quantized-2026-02-27): small, fast English
- **Parakeet** [TDT-CTC 110M](https://huggingface.co/scribekey/sherpa-onnx-nemo-parakeet_tdt_ctc_110m-en-36000) · [Unified 0.6B](https://huggingface.co/scribekey/sherpa-onnx-nemo-parakeet-unified-en-0.6b-int8-non-streaming): higher-accuracy English
- **Canary 180M Flash** [EN/ES/DE/FR](https://huggingface.co/scribekey/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8): multilingual
- **Zipformer** [Streaming EN](https://huggingface.co/scribekey/sherpa-onnx-streaming-zipformer-en-2023-06-21): live words as you speak

## Open tooling

- [scribekey-models](https://github.com/scribekey-app/scribekey-models): the open catalogue and distribution tooling behind these downloads
- [scribekey-skills](https://github.com/scribekey-app/scribekey-skills): rewrite Skills (email, notes, to-do list)

## Thanks

Built on [Sotto](https://huggingface.co/juanquivilla/sotto-cleanup-lfm25-350m) by juanquivilla, [Liquid AI's LFM2.5](https://huggingface.co/LiquidAI/LFM2.5-350M-Base), and the speech work of Useful Sensors, NVIDIA and the k2-fsa / sherpa-onnx project.

**Contact:** [founder@scribekey.app](mailto:founder@scribekey.app) · [support@scribekey.app](mailto:support@scribekey.app)