thewh1teagle/vibe

Vibe – Offline, multilingual transcription for audio & video

What it does
Vibe is a desktop application that lets you turn speech in audio or video files into text without sending anything to the cloud. It runs locally on macOS, Windows and Linux and supports a wide range of source formats (YouTube links, local files, microphone, system audio, phone recordings, etc.). The transcribed text can be exported as SRT, VTT, TXT, HTML, PDF, JSON or DOCX, and the app also offers optional summarisation, translation and speaker‑diarisation.

Key capabilities

• Offline‑first: Uses locally‑run models (Whisper, Nemotron 3.5, Parakeet TDT v3) so no data leaves the device.
• Multilingual: Works with “almost every language” supported by the underlying models.
• Multiple input sources: Files, YouTube/Vimeo/Facebook/Twitter URLs, microphone, system audio, QR‑code‑driven phone recordings.
• Export & post‑processing: Export to subtitle formats (SRT/VTT) or document formats, print directly, generate summaries via Claude API or local Ollama, translate to English, and run speaker diarisation.
• Performance: GPU‑accelerated on macOS, Windows and Linux (Vulkan, CoreML, Nvidia/AMD/Intel).
• Extensibility: Custom model URLs, CLI interface, HTTP API with Swagger docs, and configurable model arguments.

Who it’s for

• Users who need privacy‑preserving transcription (journalists, researchers, developers).
• Content creators who want subtitles or transcripts for videos.
• Anyone who prefers a graphical UI but also wants a command‑line or API hook for automation.

How to get started

1. Download the app from the GitHub‑hosted page linked in the README.
2. Install any required GPU runtime (Vulkan, CoreML, etc.) if you want hardware acceleration.
3. Open the UI, add a file or paste a video URL, choose a model, and hit Transcribe.
4. Use the side‑panel to preview, edit, export, or send the result to Claude/Ollama for summarisation.
5. For scripting, invoke the vibe CLI with --help to see available commands.

Where to find more

• Build instructions: docs/BUILDING.md
• API reference: Swagger UI served by the built‑in HTTP server.
• Community chat: Discord link in the README.
• Roadmap & issue tracker on GitHub.

Bottom line Vibe bundles several state‑of‑the‑art speech‑to‑text models into a user‑friendly, fully offline desktop app, adding useful post‑processing tools while keeping data on the user’s machine.