Privacy
Voice data is the most personal data there is. Neuron Flame treats it that way — every default keeps your audio and transcripts on your Mac. The only exceptions are features you opt into explicitly, with API keys you provide, sending data to providers you choose.
The short version
- Audio: never leaves your Mac. Period.
- Transcripts: stay on your Mac unless you turn on AI Enhancement with a cloud LLM.
- Telemetry: none. No analytics, no crash reports, no phone-home, no ping-the-server-on-launch.
- Update checks: Sparkle pings GitHub for a new version. That's the only outbound network traffic from a default install.
Where data lives
| Data | Location | Purpose |
|---|---|---|
| Recorded audio | RAM only (default) or ~/Library/Application Support/com.neuronflame.app/Recordings/ (if you opt to keep recordings) | Transcription input |
| Transcripts | SwiftData local database in App Support | History, search, retry-paste |
| API keys | macOS Keychain | Cloud LLM authentication |
| Custom dictionary & prompts | SwiftData local database | Personalisation |
| Power Mode profiles | SwiftData local database | Per-app behaviour |
| Whisper / Parakeet models | ~/Library/Application Support/com.neuronflame.app/Models/ | Local transcription |
What leaves your Mac (and when)
Default install, all toggles off:
- Sparkle update check on launch (HTTPS GET to
raw.githubusercontent.com). - Whisper/Parakeet model downloads when you click Download (HTTPS to Hugging Face / GitHub).
- Nothing else.
If you turn on AI Enhancement with a cloud LLM:
- Each transcript is sent to your chosen provider over HTTPS.
- The system + user message and the resulting reply pass between you and the provider.
- The provider's privacy policy applies to anything they retain. We can't speak for them — pick one you trust.
- Use Ollama or a local CLI if you want enhancement without anything leaving your Mac.
If you turn on Clipboard Context (off by default):
- Your current clipboard text is included in the LLM request.
- Same destination, same provider policy.
- Never written to disk.
If you turn on Window Context (off by default):
- The frontmost window's title and URL (if browser) are included in the LLM request.
- Same destination, same provider policy.
- Never written to disk.
What we never collect
- Crash reports.
- Usage analytics.
- Feature toggles you've enabled.
- Your microphone gain or mic model.
- How often you use the app.
- Your IP address.
- Anything, basically. We don't run a server.
Third-party network calls
By default, Neuron Flame talks to:
- raw.githubusercontent.com — Sparkle update feed (
appcast.xml). - github.com — when you click Download to update.
- huggingface.co — only when you download a Whisper model.
Anything else is opt-in:
- Your chosen LLM provider — only if AI Enhancement is on and configured.
- Your chosen cloud transcription provider — only if you've added one.
Deleting your data
To wipe Neuron Flame's local state without uninstalling the app:
rm -rf ~/Library/Application\ Support/com.neuronflame.app/
This removes recordings, transcripts, custom dictionary, prompts, and Power Mode profiles. Models too — you'll need to re-download.
To remove the API keys from Keychain: open Keychain Access, search for "neuronflame" or "ziva", delete entries.
Open source & auditable
Neuron Flame is open source under GPL-3.0. The full source is on GitHub. If you don't trust our description of what's happening, read the code — every network call is searchable, every persistence path is documented, every permission check is explicit.