StageWhisper processes real conversations. This page covers what data is collected, where it goes, and what controls you have.
What StageWhisper captures
During a listening session, StageWhisper captures:
- System audio from your call application (the conversation both sides hear)
- The resulting transcript (generated on your device using local speech recognition)
- AI-generated signals and insights
- Tone analysis data (if enabled on your plan)
What StageWhisper does not do
- It does not record or store raw audio files. Audio is processed on your device and discarded.
- It does not run in the background when you’re not in a session. When you stop a session, capture stops.
- It does not share data with the person on the other end of your call.
- It does not send audio to external servers. Speech recognition runs entirely on your Mac.
Local processing
Speech recognition, prompt construction, and result processing all run on your device. The desktop app handles:
- On-device transcription using a local ASR model (no cloud speech APIs)
- Building analysis prompts from your transcript, playbook, and session context
- Parsing AI responses into signals, checklists, and tasks
- Screen capture analysis and document retrieval
The backend provides authentication, session metadata storage, and AI transport. It does not process your audio or build your prompts.
Where your data lives
StageWhisper infrastructure runs in the EU. Session metadata, account data, and hosted-mode analysis results are stored on European servers.
| Data | Location |
|---|
| Audio and transcripts | Your device only |
| Screen recordings and OCR data | Your device only (encrypted at rest) |
| Reference files and indexes | Your device only (encrypted at rest) |
| Account and authentication | EU |
| Session metadata and insights (hosted mode) | EU |
| Tone analysis results | EU |
| Payment processing | Stripe (global, PCI-compliant) |
At-rest encryption
Local data stored on your device is encrypted using a per-device key stored in your macOS Keychain:
- Screen recording database (SQLCipher)
- Screenshots and compacted video
- Reference file storage, text chunks, embeddings, and summaries
Temp files containing plaintext are cleaned up when a session ends.
BYO mode and end-to-end encryption
When you connect your own AI and switch to BYO (Bring Your Own) reasoning mode, StageWhisper adds end-to-end encryption between your desktop and your AI assistant plugin. The backend is a blind relay: it passes encrypted messages between the two sides but cannot read the content.
In BYO mode, the backend can see:
- Session ID, timestamps, and connection status
- Task status (pending, approved, completed) and action type
- Envelope metadata (message IDs, content type labels)
- Error codes and retry counts
The backend cannot see:
- Transcripts
- Analysis prompts and AI responses
- Signal content and evidence
- Task descriptions, request text, and replies
- Reasoning payloads
The encryption uses XChaCha20-Poly1305 with X25519 key agreement. Keys are exchanged during the pairing process and never leave the two endpoints.
See BYO AI setup for setup details.
Screen-share privacy
The desktop overlay has a privacy toggle. When you switch to “Private” mode (it’s the default mode), the StageWhisper overlay is hidden from screen recording and screen sharing applications. This means:
- If you’re sharing your screen on Zoom, the other participants won’t see your coaching signals.
- If you’re recording your screen, the overlay won’t appear in the recording.
Toggle between Private and Visible using the button on the control panel.
AI assistant data sharing
If you connect an AI assistant, what gets shared depends on your reasoning mode:
BYO mode: All content is end-to-end encrypted between your desktop and your AI assistant plugin. The backend relays encrypted envelopes and cannot read them.
Hosted mode: Your desktop sends analysis prompts to the backend, which calls the AI model and returns results. The backend does not log prompt content or AI responses.
Raw audio is never sent to external services in either mode. See AI agents for details on the architecture.
Reference files
You can attach documents to Playbooks. See the Playbooks guide for details. Privacy specifics:
| What happens | Where |
|---|
| File storage | Your device only (encrypted at rest) |
| Text extraction and indexing | Your device only |
| Embedding model | Runs on your device |
| Excerpt sharing during sessions | Short excerpts (1-2 sentences) go to the AI for analysis, held in memory only |
| Server-side persistence | None. Excerpts are never saved on our servers |
| Telemetry and error reporting | Document content is stripped from all logs and reports |
Data retention
Your session data persists in your account until you delete it. You can delete individual sessions from the web dashboard. If you delete your account, all associated data is removed.
GDPR
StageWhisper is designed for GDPR compliance. You can:
- Access all your stored data through the web dashboard
- Delete individual sessions or your entire account
- Export session transcripts
The tone analysis pipeline measures conversational dynamics at the interaction level, not individual emotions. This means it does not generate personal emotional profiles of the other person on your call.
If you’re using StageWhisper in a context where recording consent is required by your local laws, you are responsible for obtaining that consent. StageWhisper does not notify call participants that it is running.
