I wanted to keep working with Claude Code while eating, on the treadmill in my room, or anywhere my hands are busy. Dictation tools exist, but I wanted hands out of the picture entirely: control by voice alone, and a short reply back from Claude when the task is done, all local, so I don't need another AI in the loop just to hear what Claude said. So I built Bol. Say "hey Bol", talk, pause, and the words land in Claude's box. Say "send it". Turn on talk-back and Bol reads you what Claude did when the turn ends, from a 1B model running on your Mac. Everything is local: Parakeet on MLX for recognition, a 5 MB keyword spotter for the wake phrase, Silero VAD, and a 350M model I fine-tuned to clean up fillers without dropping clauses. About 2.5 GB on first run, then nothing leaves the machine. Honest limits: macOS on Apple Silicon only, and the mic stays open while it runs. It's MIT, one line to install with uv. I'd love to hear what you'd want it to say back to you.

Bol

Control Claude Code by voice. Talk, send it, hear it back.

Control Claude Code by voice. Talk, send it, hear it back.