Whispree

What does Whispree do?

Whispree describes itself as a tool that lets you "talk to AI instead of typing". You place the cursor in any prompt input (the README names Cursor, Claude and ChatGPT), press a hotkey, and speak. Corrected text is pasted back where the cursor was, and the README says the app remembers the original focus position even if you switch windows while recording.

At the start of a recording, the README says Whispree also captures a screenshot of the focused screen and attaches it, so a vision-capable model can see what you were looking at. Speech-to-text correction runs in one of four modes: Standard, Filler Removal, Structured (for Prompt), and Custom. New recordings can be queued while a previous one is still being corrected.

Key facts

• Licence: MIT on the code, per the LICENSE file.
• Platform: macOS 14.0 or later on Apple Silicon (M1, M2, M3, M4). Microphone and Accessibility permissions are required, per the README.
• Speech providers: Groq's whisper-large-v3-turbo in the cloud, WhisperKit openai_whisper-large-v3_turbo locally (about 1.5 GB, CoreML plus Apple Neural Engine), or MLX Audio Qwen3-ASR-1.7B-8bit locally (about 1.0 GB).
• Language-model providers for correction: OpenAI through the Codex CLI OAuth token, or one of six local models via MLX (Qwen 1.7B up to Qwen Coder 30B, plus GLM-4.7 Flash 4-bit and Qwen VL for screenshot context).
• Install: brew tap Arsture/whispree && brew install --cask whispree, or a .dmg from GitHub Releases (which needs xattr -cr because the app is not notarised, per the README).
• Recording hotkey is Ctrl+Shift+R by default, with push-to-talk or toggle modes.
• Whispree registers a whispree:// URL scheme so Raycast, Stream Deck, Keyboard Maestro and AppleScript can start and stop recording.
• The app started as FreeWhisper, per the README's naming section.

What does it replace, and where does it fall short?

Whispree replaces Superwhisper and Wispr Flow for someone who dictates into AI prompt boxes on a Mac and wants a language-model cleanup step in the loop. The README says the OpenAI cleanup step runs on your existing OpenAI account through the Codex CLI sign-in, with no separate API key. It is a candidate on the open-source alternatives to Superwhisper page.

Where it falls short: Apple Silicon only, so an Intel Mac is out. macOS only, so Windows and Linux are out; for those, Yap is a cross-platform option. The DMG is not notarised, so a browser download needs a Terminal command to open. The Codex OAuth path depends on an active OpenAI account.

How does Whispree run?

The recommended install is Homebrew: brew tap Arsture/whispree && brew install --cask whispree. A signed installer is not available because there is no Apple Developer ID, so a .dmg download from GitHub Releases needs xattr -cr Whispree.app in Terminal, per the README. A build from source uses xcodegen and Xcode; SPM dependencies resolve on first build.

The app runs from the menu bar. Speech and language-model providers are picked in Settings, and models download from a Downloads pane with a "Can I Run" grade based on your Mac's RAM and chip.

Who is Whispree for?

A macOS developer with an OpenAI account who wants Wispr-style dictation into AI prompt boxes and does not want to pay for another subscription. A person on Windows, Linux, or an Intel Mac should look at Yap or Handy instead.

What limits does the README state?

The app runs on macOS 14 and later, on Apple Silicon. It is not notarised, so DMG downloads through a browser are blocked by Gatekeeper and need a Terminal command. Microphone and Accessibility permissions are required. The name section says Whispree started as FreeWhisper and was renamed for the open-source release.

Questions people ask

Is Whispree open source?

Yes. The LICENSE file is MIT, an OSI approved licence, with no additional restriction, and GitHub reports the repository's licence as MIT. Anyone can read, modify and redistribute the code under MIT terms, including for commercial use.

Does Whispree run on Intel Macs or on Windows?

No. The README says macOS 14 or later on Apple Silicon: M1, M2, M3 or M4. Intel Macs are not supported, and there is no Windows or Linux build. On other platforms, Yap and Handy both run on macOS, Windows and Linux.

Which accounts does Whispree need?

None, if you run everything locally. The README's recommended cloud setup uses Groq for speech-to-text, and OpenAI models for the correction step through the Codex CLI sign-in, which reads your existing OpenAI account credentials rather than a new API key. You can switch either step to a local model in Settings.

Can Whispree run fully offline?

Yes, if you download local models. The README lists WhisperKit and MLX Audio for speech, and six local MLX language models for correction (Qwen 1.7B, 4B, 8B, Coder 30B, GLM-4.7 Flash, and a vision model for screenshot context). The "Can I Run" screen in Settings shows which models fit your Mac's RAM and Neural Engine.

How does Whispree compare with Superwhisper?

Both dictate on macOS with a language-model cleanup step. Whispree is open-source MIT and cross-provider, so you pick speech (Groq, WhisperKit or MLX Audio) and correction (OpenAI via Codex OAuth or local MLX) yourself. Superwhisper's source code is closed. The open-source alternatives to Superwhisper page lists other options.

Sources

• Whispree README and LICENSE: github.com/Arsture/whispree, read 2026-09-28. The LICENSE file is standard MIT text.