Built this because I wanted to hear what my coding agent was doing without (a) sending agent output to a third party or (b) staring at a terminal all day.
It's a small Python daemon + macOS app that hooks into Claude Code, Codex, or anything via `heard run <command>`, and speaks intermediate output as it streams. Tool calls, status lines, failures — not just final summaries.
- Default backend is Kokoro, runs on-device, no key, no network calls for TTS
- Optional ElevenLabs if you want the premium voices
- Optional Anthropic key for in-character persona rewrites via Haiku — skip it and you get
neutral local templates
- Zero telemetry. No analytics, no crash reporters, no phone-home. Verifiable in source.
- Apache 2.0
Would love feedback as much as possible, DM or PR!
[link] [comments]