Last updated: 25 April 2026
1. What Whistle Enterprise processes
Audio recordings of meetings the user runs on their own computer, whether recorded live or imported. Transcripts produced from those recordings, generated meeting documents and any PDF, DOCX, Markdown or plain-text files exported from those documents.
2. Where files are stored
Recordings, transcripts and generated documents are kept in a local workspace on the user's computer. The workspace can be set to a local drive or a network drive. Exports are written to whatever location the user chooses at export time. There is no synchronisation to any cloud storage.
3. What leaves the device
Nothing. There is no external internet connection. The application does not initiate network connections and does not phone home.
4. Network behaviour
Whistle Enterprise runs 100% without a network connection. Recording, transcription, document generation and exporting all work the same way whether you are online or not.
Updates and previous versions are downloaded from the Whistle Enterprise website. There is no automated update functionality and no callback to a server.
5. Encryption
The local workspace can be encrypted with a password. When the password is set, recordings, transcripts and generated documents inside the workspace are encrypted at rest. The password is required to read them and without this, the contents cannot be recovered.
6. Telemetry
Whistle Enterprise sends no telemetry, no analytics events and no usage data. There is no behavioural tracking, no metrics endpoint and no background error reporting.
A user can send a diagnostic file by email. The diagnostic file does not contain any detailed information on the contents of recordings, transcriptions or generated documents. Sending it is an explicit manual action the user has to take.
7. AI models
The writing model is a custom AI model developed and fine-tuned by BlazingBanana Ltd specifically for meeting document generation, and built to run on a normal laptop CPU. The transcription model is a published speech-to-text model selected for accuracy and offline operation. Both models are bundled with the installer. Neither model is loaded from a remote service at runtime.
Content recorded, transcribed or generated by Whistle Enterprise is not used to train any model.
Questions or corrections
Anything missing from these notes, or wrong, can be flagged to [email protected]. Every message is read by a human and we aim to reply within two working days.