Privacy and local processing

Your audio stays on your machine

The panel runs inside Premiere and all analysis happens on your own computer.

What runs locally

Silence detection, transcripts and caption generation
Audio cleanup, scene and beat detection, object and face tracking, caption translation
All of it with ffmpeg and whisper.cpp, on your own computer

During setup: whisper.cpp and models (GitHub and Hugging Face), and ffmpeg if needed. The first time you translate captions, the engine and language model download from Mozilla's servers. Every download is checked against its SHA-256.
Updates: on start the panel asks Pengumi's update server (Cloudflare) for the signed release record; only standard connection data (such as your IP address) is sent. You can turn this off in Settings, and nothing installs without your click.
License: the app sends the license server a hash of the machine code and usage counters to check your license and trial/free-tier limits. Audio, transcripts and project content are never sent anywhere.
Account: if you create an account to download, your username, name, email and account status are kept on Supabase (Frankfurt); your password is stored only as a hash. The session stays in your browser, no cookie. You can delete the account from My account.

Only ffmpeg, ffprobe and whisper-cli can be run; the interface sends a tool name, not a command line.
Installer files are never run unless their SHA256 checks out.
A backup of the sequence is made before edits, by default.

Models and tools: %APPDATA%\AutoCutStudio\bin on Windows, ~/Library/Application Support/AutoCutStudio/bin on a Mac
Transcript cache: under %APPDATA%\AutoCutStudio. Clear it from Settings.
Uninstalling the plugin does not delete models.