What Is Buzz?
Buzz is a desktop transcription and translation app built on OpenAI Whisper, created by Chidi Williams and developed in the open on GitHub. The repository describes it as transcribing and translating audio offline on a personal computer, and the project site is named Buzz Captions.
The website describes two editions: "Buzz Classic", the free open-source app for Windows, Linux and macOS (Intel), and a separate native "Buzz on macOS" edition. This review covers the open-source app; the App Store link on the website returned a "not found" page on 24 October 2026, so the native edition was not assessed.
Get It
Buzz is a free download for Windows, macOS and Linux, with installers on GitHub Releases and SourceForge and Flatpak and Snap packages for Linux.
Option | How to get it |
|---|---|
| Website | buzzcaptions.com |
| macOS | .dmg from GitHub Releases |
| Windows | .exe plus two .bin files, unsigned; GitHub Releases |
| Linux | Flatpak, Snap or AppImage |
| Python | pip install buzz-captions, ffmpeg required |
| Price and license | Free, MIT (source code) |
| Privacy policy | None found; see the Preferences docs |
Version reviewed: 1.4.5, released on 23 August 2026 per GitHub Releases. The main branch has changed since (last on 17 October 2026) and already says 1.4.6, which is unreleased. On Windows, download all three files and run the .exe; the installer is unsigned, so choose More info, then Run anyway. SourceForge hosts the same installers. Intel Macs: 1.4.5 is the last version that supports them. Nvidia CUDA acceleration becomes a separate in-app install from 1.4.6.
Features Confirmed by the Sources
Everything below comes from the README, documentation and website; none of it was tested by PromptQuorum.
- Input. Import audio and video files or YouTube links, or transcribe live from a microphone; a presentation window shows live text during events.
- Whisper backends. Whisper, whisper.cpp, Faster Whisper, Whisper-type Hugging Face models or the OpenAI API, with Apple-silicon, CUDA and Vulkan acceleration where supported.
- Output. Export TXT, SRT and VTT (the website also lists CSV). A viewer offers search, playback and speed control, and a resize tool regroups subtitle length.
- Extras. Speaker identification (not on Intel macOS), speech separation before transcription, word-level timings, a watch folder, a command-line interface and plugins such as AI summary and DOCX export.
- Languages and translation. More than 90 languages per the website. Whisper translates to English by itself; other target languages need an OpenAI-compatible API, which can be a local server such as Ollama or LM Studio.
Data Flow and Privacy
Transcription runs on your computer with downloaded models, per the README and documentation. Three things can still leave the machine. Anonymous launch statistics (app version, locale, operating-system name, release and architecture) go to PostHog unless the environment variable BUZZ_DISABLE_TELEMETRY is set, and the app checks GitHub for new versions about once a week unless BUZZ_DISABLE_UPDATE_CHECK is set (available since 1.4.5).
Optional features contact an external service only when you set one up: the OpenAI API for transcription, AI translation and the AI Summary plugin (or any OpenAI-compatible server you point it to), and downloads of YouTube links. An upload address can also be configured to post live transcripts to a server.
📌Note: These statements come from the documentation and from reading the source code on 24 October 2026; PromptQuorum has not inspected network traffic. No separate privacy policy was found on buzzcaptions.com or in the repository.
What We Could Not Verify
- Accuracy and speed. PromptQuorum did not run Buzz, so word accuracy and transcription speed on any hardware are not assessed.
- Mac App Store edition. The website promotes a native macOS edition, but its App Store link returned a "not found" page on 24 October 2026, so its price, version and features could not be checked.
- Windows installer. The project says the installer is unsigned and needs three files; PromptQuorum did not scan or install it.
- Documentation versus release. Parts of the documentation describe 1.4.6 features, such as the separate CUDA install and custom whisper.cpp models, that are not in the 1.4.5 release.
- Model licences. Whisper models and any Hugging Face models you load have their own licences; PromptQuorum did not review them, and this is not legal advice.
Competitors and Alternatives
whisper.cpp
- Platforms:
- Windows, macOS, Linux
- Price / license:
- Free / MIT
- Key difference:
- C/C++ command-line engine that Buzz can use as a backend; no graphical subtitle editor
Faster-Whisper
- Platforms:
- Windows, macOS, Linux
- Price / license:
- Free / MIT
- Key difference:
- Python library that Buzz can use as a backend; aimed at developers who script it
MacWhisper
- Platforms:
- macOS
- Price / license:
- Free tier / paid Pro
- Key difference:
- Closed-source Mac app with a polished interface; Mac only
Articles about MacWhisper (2)
- MacWhisper Review: Local Speech-to-Text for MacUpdated October 3, 2026
- Meetily Review: Local, Privacy-First AI Meeting NotesUpdated September 18, 2026
Also mentioned in:
- AquaVoice Review: Voice Dictation for AI PromptsUpdated October 3, 2026
- Local Voice & Speech Tools Compared (2026): Text-to-Speech, Speech-to-Text and Voice AgentsUpdated September 20, 2026
- FunClip Review: Self-Hosted AI Video Clipping With FunASRUpdated September 19, 2026
- Willow Inference Server Review: Self-Hosted Whisper ASR and TTSUpdated September 19, 2026
whisper.cpp and Faster-Whisper are engines Buzz can run, so they complement it more than they replace it; confirm each app's current price, license and platforms on its own listing.
Frequently Asked Questions
Can I use Buzz on a computer with no internet?
Yes, according to the project FAQ: download the models on another computer, then copy the models folder (Help, Preferences, Models, Show file location opens it) to the same place on the offline machine. A download-models.py script in the repository helps prepare the cache.
Can Buzz translate into languages other than English?
Whisper's own Translate task outputs English only, and the Large-V3-Turbo model does not support it. Other target languages need an OpenAI-compatible API, which can be a cloud service or a local server such as Ollama or LM Studio.
How do I transcribe what my computer is playing?
Per the FAQ, route the application's output to a virtual audio device, such as BlackHole on macOS, VB-CABLE on Windows or PulseAudio Volume Control on Linux, and select it as the input in Buzz.
Which Whisper model should I pick?
The FAQ says smaller models are faster and less accurate, and recommends testing them on your own language. It adds that Large-V3 can sometimes invent words and that the Turbo model balances speed and accuracy.
Can I run Buzz from scripts?
Yes. The buzz add command takes files or URLs plus options for task, model type and size, language, prompt and word timestamps, and can run without opening the window by adding --hide-gui.
Verdict
Buzz is an actively maintained, free and open-source way to turn recordings into transcripts and subtitles on your own computer, with a choice of Whisper backends and a built-in editor. Against that, the Windows installer is unsigned, Intel Macs are left behind after 1.4.5, anonymous launch statistics are on by default, and speed and accuracy depend on the model and hardware, which PromptQuorum did not test. It suits people who want a graphical Whisper tool with no subscription; anyone who needs strictly offline operation should switch off telemetry and the update check and avoid the cloud options.
Sources
- Buzz Captions website — editions, features and download links, checked 24 October 2026.
- Buzz on GitHub — README, MIT license, source code and commit history, checked 24 October 2026.
- Buzz 1.4.5 release — installers and release notes, published 23 August 2026.
- Buzz documentation — preferences, telemetry and update-check variables, FAQ and installation, checked 24 October 2026.