Skip to main content
SpeakoFlow sits in your system tray and does nothing until you press a shortcut. You never switch to it to dictate. It is a layer over whatever app you are already in.

​
The dictation pipeline

Microphone

Audio is captured from the input device you picked in Settings → Audio.

Voice activity detection

Silero VAD filters out silence, so pauses while you think do not become gaps in your text.

Speech to text

A local model (Parakeet, Whisper, Nemotron, or one of the other 65 in the catalog) turns the audio into text, using your GPU or your CPU. See Acceleration.If you pick a cloud speech service instead, the audio is sent to that provider. Nothing else in the pipeline changes.

AI cleanup (optional)

When AI cleanup is on, the transcript is handed to a language model that removes filler, fixes grammar, and follows spoken edits. This runs on whichever model you chose for cleanup: the built-in engine (fully local), your own Ollama or LM Studio server, or a cloud provider.

Text into the active app

The finished text is pasted at your cursor via the clipboard, or typed directly. See Output.
With a local speech model and the built-in cleanup engine, nothing in that chain touches the network.

​
The assistant

When you ask the assistant a question (the quick ask), the question goes to whichever model you chose:
  • Built-in engine. A local llama.cpp process. No key, no network.
  • A local server. Ollama or LM Studio on your own machine.
  • A cloud provider. Your own API key for OpenAI, Anthropic, Google, or any OpenAI-compatible endpoint. See Providers.
The model gets a small set of tools (the current date and time, plus web search and screen capture if you turned them on) and decides per question whether it needs them. The loop is bounded, so a turn cannot run away. The quick ask is one question and one answer, with Copy, Insert, or Replace. Conversations are the longer-form mode: you talk, the assistant answers out loud, and you can go back and forth hands-free.

​
What stays local, what goes to the cloud

There is no telemetry, no analytics, and no account. More detail is in Privacy.
Last modified on August 30, 2026