Drive your agents by voice from ChatGPT
Your hands are on the wheel and three agents are at work. ChatGPT on your phone connects to detachbox, hears when an agent waits or finishes, and does what you say out loud: answers the agent's question, hands it the next task, starts a new terminal.

1. Connect detachbox to ChatGPT, once
At a desk, before you leave:
- In ChatGPT's settings, turn on developer mode and create an MCP app with the address
https://api.detachbox.com/mcpand OAuth. - On the detachbox page that opens, check the app's name and pick Allow access.
- In ChatGPT, let the app's actions run without asking, so a spoken answer goes through without a tap.
Assistants has every click, and the same for Claude, Grok Bot and Meta Muse.
2. Put a dot on watch
A dot is ChatGPT's always-on agent: it keeps working between your chats and writes to you first. Tell one: "Watch my detachbox terminals." It subscribes to detachbox's events and wakes up when an agent waits, finishes or stops. In its custom rules, let answer and send_message run on their own.
3. Brief it before you drive
Say once what it may decide alone:
Approve tests, builds and reading files yourself. Ask me before a push, a deploy or a delete. When an agent finishes, tell me in two sentences what it did.
4. Talk on the road
- "What are my terminals doing?"
- "Yes, let it push."
- "Tell the web project: now the mobile layout."
- "Start a Codex terminal in payments and have it fix the failing test."
- "Stop the API agent and tell it not to touch the migrations."
When something needs you, the dot writes first. Call it, answer by voice, hang up; the work it started keeps running.
5. Check what it did
Activity shows every action the assistant took for you as "You via ChatGPT", next to the ones you made yourself. To cut it off, pick Disconnect next to it on Assistants.