SparkFlow

Prove it.

Privacy claims in this market are cheap. This page is the instruction manual for verifying ours with tools you already have. No account, no trust required — watch the app's network behavior yourself.

What SparkFlow does on your machine

  1. Audio capture — the microphone stream goes straight into the local speech model. No audio is ever written to disk unless you enable history, and no audio is ever sent anywhere. Audio leaves your machine only if you deliberately export a recording.
  2. Speech recognition — Whisper and Parakeet models run locally, GPU-accelerated when you have one. No cloud round-trip exists in this pipeline.
  3. AI cleanup — this is the only step with a choice. With a local model (Ollama or llama.cpp server — a first-class option in Settings → Post Process), cleanup also runs on your machine: nothing leaves it. With a cloud provider (OpenAI, Anthropic, etc.), only the transcript text is sent — never audio.
  4. Pasting — injection into the foreground app happens through the clipboard and keystrokes. Local.
The honest caveat: if you configure a cloud cleanup provider, the dictated text goes to that provider's servers — exactly like every other AI cleanup product, except they usually also have your audio. Local mode is the zero-compromise option.

Verify it in 5 minutes

Windows

  1. Press Ctrl+Shift+Esc → Performance tab → Open Resource Monitor (bottom link).
  2. Go to the Network tab, tick the SparkFlow process, and dictate for a minute.
  3. With a local model: zero network activity while you dictate. You may see one brief update check shortly after launch (and a one-time model download the first time you add a model) — never during dictation. With a cloud provider: small, bursty HTTPS connections (the transcript text) and nothing continuous.
  4. Optional: netstat -b in an admin terminal while dictating shows every socket the app owns.

macOS

  1. Activity Monitor → Network tab → filter to SparkFlow.
  2. Dictate and watch. Local model = no traffic while dictating (an update check may appear at launch). Little Snitch (third-party) gives per-connection proof if you want it.

Linux

  1. sudo nethogs in a terminal, dictate, and watch the per-process traffic.

What "no telemetry" means here

SparkFlow phones home for exactly two things: update checks, and downloading a speech model the first time you add one. Neither carries audio or dictation text. Entering a license key never touches the network at all — keys are verified on your device. The full privacy policy says the same thing in lawyer.

← Back to the comparison