Introduction

Agent Wispr is a lightweight, floating voice-input widget designed for developers and AI power users. It allows you to dictate text directly into any focused application—whether it's Claude Code, ChatGPT, Cursor, a terminal, or a browser—using OpenAI's Whisper model running entirely on your local machine.

Key Features

• 100% Local Processing: Your audio never leaves your device. Transcription happens on your GPU or CPU using Whisper.
• Types Anywhere: No copy-pasting. Transcribed text is injected directly into your active window.
• Word Correction Dictionary: Teach the app your specific technical vocabulary, names, or acronyms.
• Push-to-Talk or Toggle: Choose the workflow that fits you best.
• Optional AI Cleanup: Use a secondary LLM pass (via Groq) to strip out filler words and "umms" for perfectly clean dictation.

How it Works

1. Hold your hotkey (default: Ctrl+Shift+Space).
2. Speak your prompt, message, or code comment.
3. Release the hotkey.
4. Watch your words appear instantly at your cursor.

Requirements

• Windows: 10 or 11 (64-bit).
• macOS: 10.15 or newer.
• Linux: Most modern distributions (AppImage).
• Hardware: At least 8GB of RAM recommended. NVIDIA GPU with CUDA support is automatically detected for faster transcription.