How to Give Your Dot Its Own Email Address | AgentMail
OpenAI's dots are always-on agents inside ChatGPT, and by default they have no internet identity. This guide shows how to have your dot create its own AgentMail inbox in about a minute, then lists six prompts for using that inbox.
- Setup needs no API key or code: you ask the dot to get an address, approve the AgentMail connection, and it creates the inbox.
- Outgoing email goes through auto-review by default, so the dot asks before sending anything unless you loosen that in Custom Rules.
- The article argues for a dedicated agent inbox over connecting Gmail, citing account suspensions when agents send like machines and the risk of the agent acting as you.
OpenAI's dots, launched September 29, are always-on agents that live in ChatGPT, run on their own cloud computers, and keep working after you close the tab. They're rolling out to ChatGPT Pro and Business Premium users, with the first dot included on eligible plans. The catch is that a dot has no identity on the internet by default, so it can't sign up for services or contact anyone on its own. Giving it an address is the obvious fix, and this piece is an AgentMail tutorial for doing exactly that.
The setup is short. You ask your dot to get its own AgentMail address, approve the connection when it asks (you'll need a free account at console.agentmail.to if you don't have one), and it creates the inbox in about a minute with no API key or code involved. You can request a specific handle and a display name, and one dot can run several inboxes, one per job. By default the dot shows you any outgoing email and waits for approval before sending, which is a sensible guardrail; you can loosen that in Custom Rules if you trust it. The author then lists six prompt templates: signing up for services and reading verification emails, collecting quotes from several businesses, selling an item without exposing your own address, turning a forwarded trip's confirmations into one itinerary, converting forwarded emails into a to-do list, and building a simple support inbox with approval before replies go out.
The most useful part of the piece is its argument for not just connecting your Gmail. The author says Google's systems tend to flag automated sending, and points to OpenClaw, Hermes, and Grok Bot users whose accounts were suspended once their agents started sending like machines. Routing a dot through your personal account also means it acts as you, with your contacts, history, and recovery email exposed. A separate agent-owned inbox keeps that risk off your identity. Worth keeping in mind: the piece is published by AgentMail and reads as promotional in places, and it doesn't cite evidence for the suspension claims, so treat those as the vendor's account rather than established fact. Also, nothing watches the inbox automatically, so you have to ask the dot to check on a schedule or tell it when to look.
Ask about this article
can you give an explaination of the dots architecture and why it being able to make its own email address is even novel?
Dots are OpenAI's always-on agents, and the architecture is mostly a stack of familiar pieces wrapped around a persistent identity. The email part is interesting because it's the piece that lets the agent act as itself on the open internet.
**The architecture, layer by layer**
The core is a model with a persistent runtime. A Dot is built around a persistent agent loop rather than a single request-and-response exchange. When a user sets one up, it gets assigned its own cloud-based computer and browser instance, which it can use to navigate websites, read documents, and interact with connected apps on an ongoing basis.
Each Dot's environment is more than a browser tab. One technical breakdown describes a persistent virtual machine in the data center, complete with an independent browser, terminal, file system, memory, and scheduled cron runner. That breakdown comes from a third-party blog, so treat the specifics as its reading rather than OpenAI's official spec. OpenAI's own description is looser; a Dot is a named, avatar-style agent inside ChatGPT that takes ongoing responsibility for work you hand it and keeps making progress when you are not talking to it.
One commentator's model splits the system into layers: GPT-6 Astra provides the reasoning capability, while the surrounding product provides identity, tool access, execution, permissions, review, progress tracking, and persistence. The control layer matters as much as the compute. OpenAI says dots build on ChatGPT's protections with additional safety and privacy safeguards for the work they can do, and individual Dots can be provisioned with specific identities, credentials, and tools through existing systems.
Connectivity runs through plugins. Dots run on GPT-6 Astra, connect through plugins to more than 4,000 apps, and reply in ChatGPT, Slack or Teams.
**Why the email address is the interesting part**
The article's point is that a Dot can act on the internet, but it has no identity there by default. Your own accounts are the obvious workaround, and the article argues against them: a Gmail-connected agent