# Voice Agent

> Hands-free spoken studio agent: opens a terminal tab with an OpenAI Realtime voice that hands work to background Claude sessions.

- Skill: `markbaindesign/voice-agent` (Agent Skill)
- Install (CLI): `npx get-skill add markbaindesign/voice-agent`
- Raw SKILL.md: https://api.skillmd.com/api/skills/markbaindesign/voice-agent/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: markbaindesign (https://skillmd.com/u/markbaindesign)
- Updated: 2026-11-15
- Page: https://skillmd.com/skills/markbaindesign/voice-agent

---


Launch the voice agent in a new Terminator tab:

```bash
terminator --new-tab -e "bash -c 'cd /media/data/dev/bain-studio && python3 studio/voice/agent.py; exec bash'"
```

Flags, added after `agent.py` when asked:
- `--full`: gpt-realtime instead of gpt-realtime-mini. Better at deciding when to start tasks, ~4x cost.
- `--duplex`: mic stays live while the agent speaks, so Mark can interrupt. Headphones only.

Then tell Mark it is running in the new tab. Ctrl+C ends it; the transcript is saved to `/media/data/Dropbox/Work/Studio/context/voice-agent/`.

All behaviour lives in `/media/data/dev/bain-studio/studio/voice/` (agent.py, tasks.py, system-prompt.md). Do not add logic here.

