← Pawscribe

Privacy Policy

Effective: June 14, 2026 · Applies to: Pawscribe 1.0.1 and later, until this document is updated · Contact: [email protected]

TL;DR

1. Local by default. Recording, transcription, and echo cancellation all run on your Mac. No network required.
2. Every cloud service is opt-in. If you don't enter an API key, not a single byte leaves your computer.
3. Zero telemetry, no analytics, no auto-uploaded crash logs. We have no server tracking what you do.

1. Who we are

Pawscribe is a local-first macOS app for meeting recording, transcription, and AI-summarization, built by Natural Emergence. It is a standalone application — there is no "user account" or "cloud database" operated by us.

2. How data flows

2.1 Data that stays entirely local by default

The following items live only on your machine. We cannot see, receive, or access any of it:

• Audio files (.m4a) → ~/Documents/Pawscribe/recordings/
• Transcripts and metadata (.json / .md) → next to the audio
• AI chat history → next to the audio
• App preferences → ~/Library/Preferences/com.naturalemergence.Pawscribe.plist
• API credentials you enter → macOS Keychain (system-level encryption)

Recordings are named pawscribe_YYYYMMDD_HHmmss.m4a. The filename itself contains no identifying information.

2.2 Local-only processing

• Local transcription via WhisperKit (open-source, models run offline on the Apple Neural Engine) or Apple's system SpeechRecognizer. Apple's on-device recognition is fully offline for supported languages; a small set of languages may be processed on Apple's servers — see Apple's own privacy documentation.
• Echo cancellation (AEC3): WebRTC algorithm running locally.
• Real-time recording uses ScreenCaptureKit for system audio + AVFoundation for the microphone, mixed in memory and written to a local file.

3. Optional cloud services (only triggered when you explicitly enable them)

3.1 AI summary / AI chat

Transcript text is sent to a provider only when you have actively configured an API key and made a request. Pawscribe is just a client — no Natural Emergence-operated server holds or routes your data. Requests go directly from your Mac to the provider.

Supported providers (you sign up, pay, and manage each one yourself):

• International: AWS Bedrock, Anthropic Claude, OpenAI, Azure OpenAI, Google Gemini, OpenRouter
• China: Alibaba Cloud Bailian (Qwen), DeepSeek, Zhipu GLM, Moonshot Kimi
• Local: Ollama (localhost) — does not leave your computer
• Custom: any OpenAI-compatible endpoint — you choose what to send and where

What gets sent: the transcript snippets you select, plus the question you type. The audio file itself is never sent to AI providers.

3.2 AWS Transcribe (optional cloud transcription, replaces local Whisper)

If you actively choose AWS Transcribe as the transcription backend in Settings, the audio file is uploaded to your own AWS S3 bucket for transcription. That is your AWS account, your bucket — Natural Emergence has no access. S3 object key prefix is pawscribe/.

3.3 Auto-update (Sparkle)

The app makes one anonymous GET request per day to https://naturalemergence.com/software/pawscribe/appcast.xml to check for updates. This endpoint is run by Natural Emergence; the standard web server log may briefly retain the request IP, User-Agent, and timestamp (typical 7–30 days) for traffic stats and troubleshooting.

We do not include any user identifier, device UUID, machine serial number, or install ID in this request. You can disable "Automatically check for updates" in Settings to stop the request entirely.

4. The "Never leave local" hard switch

Settings → Privacy → Never leave local. Once enabled:

• All cloud AI calls are refused outright (no request issued)
• AWS Transcribe falls back forcibly to local Whisper
• Stored credentials are not deleted — turn the switch off and you're back online
• A 🔒 "Local" badge stays visible at the top of the recording popover

Useful for confidential meetings, sensitive client information, medical or legal conversations, etc.

5. System permissions

Pawscribe requests the following macOS permissions on demand. Each is a system dialog you can revoke any time in System Settings → Privacy & Security.

• Microphone: records your voice. Refusable (mic track unavailable).
• Screen Recording: captures the system audio output — note: we do not read screen pixels. Refusable (system audio track unavailable).
• Speech Recognition: Apple SpeechRecognizer for live captions. Refusable (use WhisperKit instead).
• Notifications: recording start / stop banners. Refusable.

6. What we do not collect

• ❌ Behavioral analytics, click tracking, A/B test data
• ❌ Auto-uploaded crash logs
• ❌ Device fingerprints (IDFA / serial number / MAC address / hardware UUID)
• ❌ Recording content or transcription content, in any form
• ❌ The API keys you enter
• ❌ Account sign-up (the app has no account system)

The only outbound network requests come from you: ① the AI / transcription providers you configure, ② the auto-update check (which can be turned off).

7. Third-party components

The app embeds the following open-source components, which run on your device and do not phone home:

• WhisperKit (MIT) — local speech recognition
• FluidAudio (MIT) — speaker diarization
• WebRTC audio processing (BSD-3-Clause) — echo cancellation
• swift-markdown-ui (MIT) — Markdown rendering
• KeyboardShortcuts (MIT) — global shortcuts
• Sparkle (MIT) — auto-update

8. Children

The app is not designed or marketed to children under 13, and does not knowingly collect information from children.

9. Changes to this policy

Material changes will be announced in the app changelog and reflected in the "Effective" date at the top of this page. We do not track who has read this document.

10. Contact

Questions about this policy or your local data? Email [email protected].

This policy describes the full data handling behavior of the standalone Pawscribe application. If anything here disagrees with the app's actual behavior on your machine, the app's behavior wins — please email us so we can correct this document.