100% Offline & Local-First Architecture

Privacy Policy

Last updated: September 20, 2026. Verbtm is designed from the ground up so you never have to trust us with your voice or personal data.

Our Core Commitment: Zero Cloud Speech Processing

When you speak into Verbtm, your microphone audio is captured in volatile memory (RAM), transcribed locally on your GPU or CPU using open-weights Whisper models, and injected directly into your active window. Your audio, keystrokes, and transcriptions never leave your physical device. We operate no speech-processing servers and store zero user recordings.

1. What We Never Collect

By design, Verbtm does not harvest, store, transmit, or monetize personal information. Specifically:

Voice & Audio Streams: Audio buffers exist only in temporary RAM during active dictation and are immediately released after inference. No audio files are recorded or uploaded.
Cloud Speech Logging: Your spoken audio and dictation are never transmitted to any cloud server, third-party API, or external database. All transcription processing occurs 100% locally on your machine.
Clipboard Contents: Verbtm never reads, writes, replaces, or inspects your operating system clipboard history.
Keystroke Logging: Verbtm listens globally only for your configured push-to-talk hotkey combination. It never monitors, logs, or transmits keystrokes typed elsewhere.
Accounts & Biometrics: Verbtm requires no user accounts, passwords, email sign-ups, or voiceprint profiling.

2. Network Activity & Outbound Requests

Verbtm is engineered to run completely offline once set up. The application only establishes outbound network connections in three specific, transparent circumstances:

A. Model Weights & CUDA Runtime Downloads (Initial Setup)

When installing or choosing a new transcription model, Verbtm downloads public Whisper model weights (from Hugging Face CDN) and optional NVIDIA CUDA execution packages (from Cloudflare R2). These are standard HTTPS downloads of static binary files; no personal data or device identifiers are sent.

B. Software Update Inquiries

To check if a newer version of Verbtm is available, the desktop client requests a static JSON file (https://verbtm.app/version.json). This request sends only standard HTTP headers (OS platform and generic user-agent). It contains no unique user IDs, license tokens, or telemetry.

C. Voluntary Feedback & Bug Reports

If you voluntarily submit feedback via our web portal (verbtm.app/report), we receive only the details you choose to include in the form (e.g. summary, description, and optional contact email).

3. Local Storage & Diagnostic Logs

Verbtm stores configuration files and diagnostic logs strictly on your local disk:

Windows %APPDATA%\Verbtm\config.json %APPDATA%\Verbtm\verbtm.log

macOS ~/Library/Application Support/Verbtm/config.json ~/Library/Application Support/Verbtm/verbtm.log

Notice on Diagnostic Logging & Transcribed Text

For local latency benchmarking, performance tuning, and troubleshooting, the local verbtm.log file records runtime events (device sample rates, model load times, hotkey states, and crash traces) alongside the resulting transcribed text snippets and timing metrics (for example: [Transcriber] Completed in 142ms: "...").

100% On-Device & Zero Cloud Transmission: This log file resides strictly on your local disk and is never transmitted, synced, or uploaded to any remote server or third party.
Bounded Rolling Backups: The log file is strictly size-capped at 5 MB with up to 3 rolling backups (max ~20 MB total disk footprint), automatically purging older entries over time.
Full User Ownership & Control: You can inspect, edit, or delete verbtm.log at any time. When submitting feedback via /report, attaching log excerpts is entirely optional and manually reviewed by you.

4. Website & Cookies (`verbtm.app`)

Zero Tracking Cookies: Our website uses zero marketing cookies, zero cross-site trackers, and zero behavioral analytics suites (such as Google Analytics or Meta Pixel).
Edge Caching & Security: The website is served via Cloudflare edge infrastructure. Standard server request logs (IP address, user agent, timestamp) are retained transiently for DDoS defense and edge routing.
Cloudflare Turnstile: On the feedback portal (/report), we utilize privacy-focused Cloudflare Turnstile to prevent bot abuse without tracking users across the web.

5. GDPR & CCPA Compliance by Design

Under regulations such as the General Data Protection Regulation (GDPR) and California Consumer Privacy Act (CCPA), individuals have rights regarding access, deletion, and portability of personal data.

Because Verbtm operates on a strict principle of Data Minimization and retains no user audio, transcripts, or personal data on remote systems, we have no personal data to sell, disclose, or delete. You retain 100% ownership and physical custody of all data on your computer.

6. Questions & Contact

If you have questions about this Privacy Policy or Verbtm's offline architecture, please contact us:

Feedback & Support Portal support@verbtm.app