Privacy Policy

Codescribe is designed with privacy as a core principle. Your audio is processed locally by default. This page restates the product’s privacy truth for buyers and users; the engineering source of record is docs/guide/privacy.md in the repository.

Privacy summary

Data	Where it goes	Your control
Audio	Your Mac only	Never leaves unless you enable cloud STT
Transcripts	Your Mac only	Stored under ~/.codescribe/transcriptions/
AI requests	Your chosen provider	Only if AI formatting / agent features are enabled
Telemetry	Nowhere	No tracking, no analytics, no phone-home

What stays local

These stay on your machine and cannot be flipped “to the cloud” by us:

• Whisper model — runtime-resolved local path/cache; local STT runs on-device.
• Audio recording — RAM only; deleted after transcription (unless you explicitly enable debug audio dumps).
• Transcripts — ~/.codescribe/transcriptions/; you control retention.
• Configuration — settings plus optional ~/.codescribe/.env.
• API keys — macOS Keychain (and/or restricted local config).
• Prompts — ~/.codescribe/prompts/.

No network is required for recording audio, running Whisper transcription, pasting text, storing transcripts, or hotkeys. Codescribe works completely offline in Raw mode.

What can leave your Mac (opt-in only)

Optional AI formatting / agent calls

When you enable AI formatting or agent features, transcribed text (not audio) is sent to the provider you configure over HTTPS. That may be OpenAI-compatible Responses/chat endpoints, Anthropic, a self-hosted stack such as LibraxisAI, or local Ollama. We do not send audio files, unrelated transcripts, or system inventory as part of those requests.

Optional cloud STT

If you turn off local STT and configure a cloud STT endpoint, audio may be sent after capture for a final transcript. Live preview still stays local in the current product shape.

Cloud is never required for core dictation. Maximum-privacy setups disable AI formatting, force local STT, and leave LLM endpoints unset.

System permissions (TCC)

macOS may ask for:

• Microphone — record speech while you hold a hotkey or start a session.
• Accessibility — paste text and interact with the focused app when you ask it to.
• Input Monitoring — detect modifier/hotkey state (not a keylogger of content).
• Speech Recognition — when using Apple’s speech path for live letter-level transcript.

We do not request camera, location, contacts, or blanket file access outside the app’s data directory (except optional full-disk flows you deliberately enable).

Tracking and analytics

No tracking, no analytics, no phone-home. The product does not ship third-party analytics SDKs or silent telemetry beacons. Network activity happens only when you configure and use a remote provider (or download a release from GitHub).

Data retention and rights

Transcript history lives under ~/.codescribe/. You can disable history, delete files, or remove the whole directory. For EU users under GDPR-style access/erasure/portability requests: your data is already on your Mac as plain files you control. We do not operate a cloud account that holds your dictation history by default.

Who we are

Codescribe is built by VetCoders / LibraxisAI. Questions: hello@vetcoders.io · Support · source at github.com/vetcoders/codescribe.