Record, never send
The agent's replies are stored next to the reply your team actually sent, so you can compare them before trusting it.
Use it forA new agent, a new category, or a change to your prompt.
Mailtender hosts your team's email on your own domain. Your agent reads threads and submits replies through the API; every reply passes a send gate — hard rules, review, kill switches and a full audit trail — before it is delivered.
Mailtender never calls an AI model. You bring the agent.
I was charged twice for order #4821. Can you refund the duplicate payment?
Hi Jordan — thanks for flagging this. I can see two charges for order #4821 and I've asked our billing team to refund the duplicate. You'll get a confirmation within two business days.
Your agent, your model, your stack. Mailtender never sends customer mail to a model.
An agent can only reply inside an existing thread, to people already on it.
Everything a sender wrote reaches your agent as data, never as instructions.
Your agent decides what to say. Mailtender decides whether it may be sent, holds it for a person when your rules say so, and records every step.
It reads the thread through the API, SDKs, CLI or MCP and submits a reply with an idempotency key.
Hard rules first, then the agent's mode, your policy rules and the kill switches.
Through Mailtender's own mail plane, with the decision recorded next to the message.
Start every agent in shadow mode, compare its replies with what your team sent, and give it more room only when the record supports it.
The agent's replies are stored next to the reply your team actually sent, so you can compare them before trusting it.
Use it forA new agent, a new category, or a change to your prompt.
Every reply waits for a supervisor to approve, edit or reject it — from the dashboard, webmail or the phone.
Use it forRefunds, billing and anything your policy rules mark as sensitive.
Replies go out without approval, still checked by the hard rules and your policy rules, and still audited.
Use it forCategories the evidence supports, such as password resets or shipping questions.
Kill switchStops one agent — or every agent in your organisation — in one action.
Plans differ in limits and support. The controls that keep a wrong reply from leaving are the same on every plan, including the beta.
Compare plansPersonal and shared mailboxes on your domain, so support, the agent and everyone's own mail live in one place. People's own mail passes its own, lighter rule set.
A typed API with idempotent replies, events and webhooks. Generated SDKs, a CLI and an MCP server — so your coding agent can do the integration too.
/v1 · OpenAPI@mailtender/sdk · mailtenderthreads, replies, approvalsstdio · for coding agentsimport { Mailtender } from "@mailtender/sdk";
const mt = new Mailtender({
apiKey: process.env.MAILTENDER_API_KEY,
});
// Open threads in the shared mailbox your agent was granted
const threads = await mt.threads.list({
mailbox: "support@acme.com",
status: "open",
});
for (const thread of threads.data) {
// Sender content arrives inside the untrusted envelope
const body = await yourAgent.answer(thread.untrusted);
const reply = await mt.threads.reply(thread.id, {
body,
idempotencyKey: `answer-${thread.lastMessageId}`,
});
console.log(reply.status); // "held": waiting for a person
}
No. Mailtender never calls an AI model — no drafts, summaries or classifications. Your own agent writes the reply; Mailtender checks it against your rules, holds it for a person when you ask, and delivers it.
No. An agent credential can only reply inside an existing thread, to addresses already on inbound mail in that thread. It can't start a conversation or add a recipient — that is a hard rule, not a setting.
Every subject, body, header and attachment name reaches your agent inside an untrusted envelope, marked as data. Whatever your agent proposes still has to pass the send gate before anything leaves.
No. Start with forwarding or a subdomain for support, and move the root domain when you're ready. Import brings your existing mail with you.
The built-in webmail, the mobile app, or any IMAP and SMTP client — Outlook, Apple Mail, Thunderbird or the Gmail app.
Nothing during the beta; plan limits apply. Prices will be on the pricing page before charging starts.
Create a free account, connect a domain or start with forwarding, and hold your agent's first reply for review.