YOUR WORKSPACE / YOUR CHOICES
A closer look
at privacy.
Buddy is built around a local workspace. Cloud speech and AI features are optional and have separate permissions.
This website and browser demo
The website’s product examples use fictional meeting content. Its sample questions and setup tabs demonstrate the interface; they do not contact AI providers or change your app settings. The separate live voice demo does contact Deepgram after you start it.
The embedded interactive app demo opens a separate, temporary sample workspace. Edits stay in memory and reset when you close or reload it; it does not read or change your saved workspace. Its frame does not allow microphone or camera access. The separate full browser workspace saves notes and imported document excerpts in your browser’s storage. You can clear that workspace from its Privacy settings. Browser recording uses your microphone after you start it and grant permission. It does not create a live transcript.
This page does not install advertising analytics, marketing trackers or a mailing-list form. GitHub Pages hosts the public website. The optional live voice gateway runs on ChatGPT Sites / Cloudflare. These platforms process connection metadata needed to serve the website and demo.
The live voice demo
Pressing “Try live transcription” and granting microphone permission sends up to 45 seconds of microphone audio through Buddy’s demo gateway to Deepgram Nova-3. The gateway keeps the provider API key on the server. It does not write your audio or transcript to a database or file, and does not log their contents. Your transcript exists in the current tab; it disappears on refresh.
Every demo speech request includes Deepgram’s mip_opt_out=true option, excluding it from the provider’s Model Improvement Program. Deepgram explains the opt-out here. Transport and service-request metadata may still be processed by the hosting and speech providers. This is cloud processing, not end-to-end encryption.
The demo stops automatically after 45 seconds, when you stop it or when the page becomes hidden. Use non-sensitive sample speech. The public demo has a durable budget of 20 sample starts per UTC day, three per network per day, and 100 total. Failed speech connections can also count. To enforce these caps, the gateway stores a random session ID, a daily keyed hash of the connection address, a date and a used flag; it does not store the address itself. These limited quota records remain until the demo budget is retired. Shared networks may share a limit. Audio and transcript content is not stored or logged by the application.
The GitHub stars badge requests the public repository star count from GitHub. It sends no meeting content, provider key or audio.
The native Buddy workspace
Notes, transcripts and indexed source excerpts are saved on your device. Native audio is kept separately from the notes database, and retaining audio files is optional. Device workspaces do not currently sync with one another.
You choose on-device speech recognition or a configured cloud speech provider. Cloud transcription sends the recording audio you authorize. Cloud questions use selected context from saved meetings; including imported files in cloud answers requires a separate Settings permission, off by default.
Provider keys are stored using the operating system’s credential store. No provider keys are included in browser JavaScript or the downloadable browser workspace. The live voice demo uses a server-side secret. Provider services have their own terms, processing practices and usage limits.
Model training and your choices
Buddy does not train its own models on your meeting content. On-device speech stays on your device. Native Deepgram transcription now requests the same model-improvement opt-out as the website demo. With your own provider keys, account terms and settings also apply.
We do not currently promise “no provider training” across every AI feature. Voyage documents an account opt-out for training and retention; its status cannot be inferred from an API key. NVIDIA’s development endpoints have their own terms and restrictions. Buddy-managed AI is planned, and its final privacy and retention commitments will be published before launch.
Your controls
Start and stop recording yourself. Let other participants know before recording. You can edit your notes, remove saved sources, export a meeting, revoke cloud permissions or clear the local workspace in Buddy’s Settings.
macOS and iPhone native versions are still undergoing platform and device acceptance. This website is an early-access introduction and a browser demo, not a claim that every native feature has been verified on every platform.
Back to Buddy