Speak to your cursor: open-source voice input for AI prompts

What it does

OpenLess is a cross-platform voice-input app for macOS and Windows. Hold a global hotkey, speak, and it transcribes, polishes, and inserts the result at your cursor in any text field — ChatGPT, Cursor, Notion, or a plain email draft. If the target application blocks synthetic keystrokes, it falls back to the clipboard so nothing gets lost. It aims to replace subscription tools like Typeless and Wispr Flow while keeping your audio and dictionary data on your machine.

The interesting bit

Its “Structured” mode does not merely transcribe: it reformats your rambling into a tight, context-rich prompt ready for ChatGPT, Claude, or Cursor. A built-in Style Pack Marketplace lets you switch between custom system prompts — from terse commit messages to formal reports — via hotkey, and publish your own for others to install.

Key highlights

• Streams polished text character-by-character to the cursor as it generates, with an automatic one-shot paste fallback for apps that reject streamed input.
• Supports both cloud ASR (Volcengine streaming, Apple Speech, Whisper-compatible batch) and local ASR (bundled Qwen3-ASR, Windows Foundry Local Whisper).
• Polishing layer works with Ark, DeepSeek, OpenAI, Doubao, Anthropic, or any OpenAI-compatible endpoint you supply.
• Includes a separate “Selection-ask” hotkey that opens a floating panel to run voice Q&A against highlighted text in any app.
• Multilingual UI and a translation hotkey that converts speech directly into a configured target language.

Caveats

• macOS builds are ad-hoc signed and currently trigger Gatekeeper warnings that must be manually overridden before the app will launch.
• Linux appears in repository topics, but the README only documents macOS and Windows support, leaving Linux availability unclear.
• Every dictation is an independent request; the app deliberately does not accumulate conversation context or act as a conversational agent.

Verdict

Try it if you dictate prompts, emails, or specs all day and would rather own your pipeline than rent it. Skip it if you need a conversational assistant or an iOS client — this is strictly speech-to-cursor, not speech-to-assistant.