ghost pepper

what it does

ghost pepper is a macOS hold-to-talk speech-to-text app running entirely on local models. press a key, talk, release — text appears. no cloud, no API calls, no data leaving your machine.

why it matters

voice input was the last gap in the local-first AI stack. LLMs run locally, embeddings run locally, but voice still bounced through someone else’s servers. ghost pepper closes that gap.

the creator is already using it as a voice interface for their other agents — not just dictation, but “speech to agent.” the input layer going local before the inference layer.

key features

• 100% local → no data leaves your machine, ever
• hold-to-talk → natural push-to-talk interface, like a walkie-talkie for your AI
• MIT license → fully open source
• agent-ready → designed to pipe voice input into agent workflows

self.md relevance

once your voice never leaves your machine, the entire personal AI pipeline can run air-gapped. ghost pepper is the front door to offline agents. pair with local TTS (Voxtral) for full duplex voice without touching a server.