https://github.com/shintemy/talky
Speak it. Ship it. — The first fully-local voice engine for vibe coding. On-device Whisper + LLM, zero cloud.
https://github.com/shintemy/talky
apple-silicon llm macos ollama productivity python qt speech-to-text voice-input whisper
Last synced: 2 months ago
JSON representation
Speak it. Ship it. — The first fully-local voice engine for vibe coding. On-device Whisper + LLM, zero cloud.
- Host: GitHub
- URL: https://github.com/shintemy/talky
- Owner: shintemy
- License: mit
- Created: 2026-03-12T14:26:35.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2026-04-13T10:38:06.000Z (3 months ago)
- Last Synced: 2026-04-13T12:20:55.877Z (3 months ago)
- Topics: apple-silicon, llm, macos, ollama, productivity, python, qt, speech-to-text, voice-input, whisper
- Language: Python
- Size: 1.27 MB
- Stars: 6
- Watchers: 0
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Talky
**Speak it. Ship it.** The first fully-local voice engine for vibe coding.
On-device Whisper + LLM — hold to talk, release to paste. No cloud. No compromise.
**Language:** English | [中文](README.zh.md)
## Download
**[Download Talky DMG (latest release)](https://github.com/shintemy/talky/releases/latest)**
> macOS Apple Silicon · Unsigned / not notarized — see install steps below.
## How It Works
1. **Hold** the Fn key to start recording
2. **Speak** naturally
3. **Release** — Talky transcribes your speech (Whisper) and refines it with a local LLM (Ollama)
4. **Text is pasted** into the active app automatically
If no input field is focused, a floating copy panel appears so you can manually paste.
## Install
1. Download the `.dmg` from [Releases](https://github.com/shintemy/talky/releases/latest)
2. Open the DMG, drag **Talky** to **Applications**
3. First launch: macOS will block the app — go to **System Settings → Privacy & Security** and click **Open Anyway**
4. Grant **Microphone** and **Accessibility** permissions when prompted
5. The setup wizard will guide you through installing [Ollama](https://ollama.com/download), an AI model, and a speech model
## Features
### Vibe Coding Mode
Switch to **Vibecoding** in Dashboard → Configs to turn spoken ideas in any language into concise English prompts — ready to paste into Cursor, Claude, or any AI coding agent.
- **100% Local** — ASR + LLM run on your Mac, nothing is uploaded
- **Hold-to-Talk** — simple press-and-hold interaction, no wake words
- **Smart Paste** — auto-pastes into the active text field; fallback copy panel when no target
- **Custom Prompt** — edit the LLM system prompt to match your writing style
- **Dictionary** — add custom terms for better recognition of names, jargon, etc.
- **Multi-Mode** — Local / Remote Ollama / Cloud server
- **Daily History** — all transcriptions archived locally
## Requirements
- macOS (Apple Silicon: M1 / M2 / M3 / M4)
- [Ollama](https://ollama.com/download) installed
- ~10 GB free disk space (for AI + speech models)
## Troubleshooting
**App won't open?**
System Settings → Privacy & Security → scroll down → click "Open Anyway".
**No text output?**
Make sure Ollama is running (`ollama serve` in Terminal).
**Whisper model issues?**
Delete `~/.cache/huggingface/hub/` and re-download through the app.
## License
Private project. All rights reserved.