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
Download Talky DMG (latest release)
macOS Apple Silicon · Unsigned / not notarized — see install steps below.
- Hold the Fn key to start recording
- Speak naturally
- Release — Talky transcribes your speech (Whisper) and refines it with a local LLM (Ollama)
- Text is pasted into the active app automatically
If no input field is focused, a floating copy panel appears so you can manually paste.
- Download the
.dmgfrom Releases - Open the DMG, drag Talky to Applications
- First launch: macOS will block the app — go to System Settings → Privacy & Security and click Open Anyway
- Grant Microphone and Accessibility permissions when prompted
- The setup wizard will guide you through installing Ollama, an AI model, and a speech model
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
- macOS (Apple Silicon: M1 / M2 / M3 / M4)
- Ollama installed
- ~10 GB free disk space (for AI + speech models)
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.
Private project. All rights reserved.