Voxlert gives each session a distinct character voice. Not pre-recorded sound effects — a light LLM writes an in-character phrase for each event, local TTS voice-clones the character speaking it, then audio filters (pitch, reverb) match the character's signature sound.

So when tests fail, SHODAN doesn't play a chime — she says "Pathetic. Your test suite has failed." When a build finishes, the StarCraft Adjutant says "Transmission received — build process complete." Each phrase is unique and describes what actually happened.

How it works: CLI hook fires on Claude Code / Cursor event → light LLM generates phrase → TTS voice-clones character (Qwen3-TTS on Apple Silicon, Chatterbox on CUDA) → audio effect filters applied. Fully local TTS, LLM via OpenRouter (fractions of a cent) or local Ollama.

Install: npx @settinghead/voxlert --onboard

Works on macOS (Apple Silicon). Linux support in progress.

Demo: https://youtu.be/5xFXGijwJuk

Happy to answer questions — especially about the local TTS setup, which is the main friction point.