Questions
The ones people ask first. Answered plainly, including what is not there yet.
- Which agents does it run?
- Claude Code and OpenCode were tried for real. Any agent that runs in a terminal can be added with one entry in a file: its name, its command, how to install it.
- Does my code or my mail leave the machine?
- They stay on it. What an agent reads is sent to the AI model that agent uses, as it would be on your own computer. If you turn on answers in your place, the screen of that conversation is sent to the models named in your settings.
- Can an agent read the keys I give it?
- A key you hand to an agent is in its machine: that agent can read it, and no other. The key that sends code out and the password of a mailbox never enter any machine.
- What stops an agent that goes wrong at night?
- It can only touch what you attached to write. Nothing leaves without you. When it asks, several models must all agree, under a charter you wrote; otherwise it stops, and your phone is told. A limit of answers in a row, and a log of each.
- Does it send mail?
- No. A mailbox is read, never written. An agent drafts a reply as a file; you send it from your own mail.
- Which mailboxes work?
- Those that speak IMAP with a password. It was tried against a mail server of ours, not yet against one of the large providers.
- How do agents work together?
- Through a folder you gave to both. One leaves a request there; the other is woken by its arrival and does it. They need not be the same tool.
- Can something I already run start an agent?
- Yes. A wake-up can have an address of its own: a scheduler, a workflow tool or a script calls it, and the line you wrote is typed to the agent. The address is shown once.
- Do I have to read a terminal?
- No. Each conversation has a simple screen: what is ready to read, and what the agent says now, as plain text. An agent can also be given six verbs of its own (tell, ask, show, read, status, doing): you then talk with it as in a messaging app, with buttons for its questions, and one line saying what it is doing while it works. Tried for real with OpenCode. Its terminal is one tap away for whoever wants it.
- Does it work on a phone?
- The page is the same on a computer, a tablet and a phone, and installs as an app. On an iPhone or iPad, alerts come once it is installed that way.
- What does it cost?
- Nothing: it is open source, under the MIT licence. You pay what you already pay for your agents and models.
- Can several people use it?
- Several operators can sign in. A conversation can also be handed to someone with a code, without an account, and taken back.
- How is it different from tmux, a container, or the tool's own remote control?
- A machine of its own for each agent, with only what you attached; keys that send things out kept outside; wake-ups, answers by agreement and the send step already there. The security page shows a real trial.
- What is not there yet?
- Models running on the machine itself. A voice box.