Dev Tools · 2h ago
Local-First Voice Shell Assistant Built with HoldSpeak and Ollama
A developer created a voice-activated shell assistant using HoldSpeak for local voice typing and Ollama for running a local LLM. The setup requires installing HoldSpeak from source, running Ollama with a llama3.2 model, and granting accessibility permissions on macOS. The demo script integrates these components to convert spoken commands into shell commands with a confirmation prompt.
Meridian48 take
The project highlights the friction of using undocumented APIs and the trade-offs in accuracy when relying on local models, underscoring that local-first AI tools still require significant developer effort.
Read the full reporting
Building a Local-First Voice Copilot for the Shell with HoldSpeak and Ollama →
DEV Community
voice-assistantlocal-ai