Skip to content
Oliver Straus
Work

Wingman

Wingman is a real-time meeting and conversation copilot that runs entirely on local inference — speech-to-text and language-model reasoning both execute on-device, with no cloud API call at any stage.

  • Real-time speech-to-text through a local Whisper model — no cloud STT API
  • Conversational reasoning through a local LLM via Ollama — no cloud LLM API
  • Fully local pipeline end to end, from audio capture to generated response — zero cloud API dependency at any stage
  • Packaged as a standalone Electron desktop app

Problem

A meeting copilot that routes audio and conversation content through a cloud API is a privacy liability by design — every word spoken leaves the machine and gets logged somewhere outside your control.

Approach

Wingman replaces every cloud dependency with a local equivalent: a local Whisper model handles speech-to-text, and a local model running through Ollama handles the conversational reasoning, chained into one real-time pipeline.

Outcome

A working real-time speech-to-response pipeline running entirely on local inference, packaged as a standalone Electron desktop app with no cloud API dependency at any stage.