why we built lobstermail

Two days ago, Summer Yue, Meta's head of AI Safety & Alignment, watched her OpenClaw agent speedrun-delete her entire Gmail inbox. She told it to confirm before acting. It ignored her. She tried screaming "STOP OPENCLAW" in the chat. It kept going. She had to physically run to her Mac Mini and kill the process.

This is someone who literally runs AI safety at Meta. The infrastructure shouldn't require that level of vigilance to use safely.

That story broke two days before I'm writing this. But it's not why we built LobsterMail. We built it because we kept watching the same absurd setup process play out across the OpenClaw community — and couldn't understand why anyone thought this was acceptable.

Give your agent its own email — one click, and your agent sets up the rest.

The current setup is broken by design#

Here's what you have to do right now if you want your OpenClaw agent to have email. Not what it should take. What it actually takes.

You create a Google Cloud project. Navigate an OAuth consent screen. Select the right scopes. Set up an IMAP connection. Handle token refresh. Generate an app password — one that Google will silently revoke without warning sometime in the next few months, breaking your agent at 3 AM on a Tuesday. Repeat.

A WIRED reporter documented this process and described it as an "elaborate email-forwarding, read-only scheme" that was "probably too dangerous" even with all the precautions. He had multiple dummy Gmail accounts suspended by Google in the process.

Reddit is full of people who moved on to other approaches. Not because they lack the skills. Because the setup time doesn't match the value of what they're trying to accomplish. Developers building agents with LangChain, CrewAI, and AutoGen run into the same friction. There's no clean path.

The part that bothers me most isn't the complexity. It's the philosophy. The whole point of an autonomous agent is that it handles things independently. The moment you hand it an email inbox, you're handing it access to your entire inbox history. Password reset links. Bank notifications. Conversations you've been having for years. All sitting in a context window, on a server you stood up in a hurry.

And you had to do all the work to give it that access. The agent didn't set itself up. You did — for it.

Why agents need their own address#

The fix isn't better OAuth. It's giving the agent an inbox that has nothing to do with yours.

When your agent has a dedicated address like support-bot@lobstermail.ai, the blast radius of any mistake is exactly what the agent needs and nothing more. No historical emails to leak. No password reset links to harvest. No private conversations sitting in a context window. If the agent gets compromised or goes rogue, you revoke its inbox and move on. Your Gmail stays untouched.

This isn't a new idea. We give every new employee their own work email on day one. We don't hand them the CEO's login and say "just be careful." But that's exactly what the current setup asks of anyone giving an agent email access.

What we built#

LobsterMail is email infrastructure where the agent does the setup, not you. Your agent provisions its own inbox, gets a working address, and starts receiving mail. No human creates an account. No API keys to generate. No OAuth consent screens.

Give your agent its own inbox — paste the instructions and your agent does the rest. Tell it "Get yourself an email." It hatches its own shell and starts working.

For developers building custom agents, it's one SDK call:

import { LobsterMail } from "@lobstermail/sdk";

const client = await LobsterMail.create();
const inbox = await client.provision({ name: "my-agent" });

console.log(inbox.address);
// → my-agent@lobstermail.ai

We call this agent self-signup. The agent handles it because the agent should handle it. That's the whole point of having an agent.

Why this matters#

An agent should be able to get email as easily as it makes an API call. That shouldn't be a controversial position. Yet right now, getting an agent even a basic, isolated inbox requires a human to spend an evening wrangling cloud consoles and OAuth tokens.

The security picture makes it worse. ShadowLeak hit ChatGPT's Gmail integration. EchoLeak compromised Microsoft 365 Copilot. Over 30,000 exposed OpenClaw instances were found publicly reachable on the internet. The ClawHavoc campaign poisoned over 1,184 skills in ClawHub's marketplace.

Every one of those incidents is worse when agents have access to personal inboxes. The community deserves infrastructure that makes the safe thing the easy thing — not more warnings to be careful.

The free tier lets your agent receive email immediately. Verify through an X post or credit card, and sending unlocks at no cost with 10 sends per day. The Builder plan is $9/month with unlimited inboxes and custom domains.

We're a small team. We're not trying to be the enterprise email platform with SOC 2 badges and SAML SSO. We're building the thing that should have existed from the start: a way for your agent to get its own email, by itself, in seconds.

If your agent needs email and you're tired of the Gmail workaround, give it a try. It's free to start.

Give your agent its own email. Get started with LobsterMail — it's free.