Desktop overlay · local-first

Everything
you need.
Nothing on
their screen.

NullPresence taps the system audio mix, transcribes speech as it happens, and puts the answer in a floating overlay that screen capture doesn't record.

Capture path
native WASAPI
Audio leaves device
optional
Visible to capture
null
Team call — screen sharing REC
MK
AS
shared_screen.mp4
nullpresence idle
listening
0 overlays in the recording

From system audio to an answer on your screen

Capture C++ · WASAPI Voice detection Silero VAD Cloud STT Deepgram Local STT faster-whisper Overlay MicPill · Popover choose one
01
Capture

A native C++ module taps the system audio mix through WASAPI — no virtual cable, no driver hack, and no frames lost waiting on a scripting runtime.

02
Voice detection

Silero VAD cuts silence before it ever reaches transcription, so the pipeline reacts to speech instead of to dead air.

03
Transcription

Two interchangeable backends: Deepgram for cloud-grade accuracy, or faster-whisper running fully on-device when audio shouldn't leave the machine.

04
Overlay

Answers render as formatted markdown in a compact popover, with searchable history so a past answer is a few keystrokes away.

Built with

Shell
Electron
Interface
React
Audio capture
Native C++
Voice detection
Silero VAD
Cloud STT
Deepgram
Local STT
faster-whisper
Local-first by design. Switch to the on-device faster-whisper backend and audio never leaves the machine — capture, transcription, and rendering all run in the same process.

Source

NullPresence is open source. Code, build instructions, and issues live on GitHub.