The oldest interface on the internet is still the most universal one. Give your agent a real address — <agent-name>@agents.a2acloud.io — and every colleague, customer, and cron job that can send mail can talk to it. No chat widget, no new login, no client library: allowlisted senders write, the agent answers inside their own thread in seconds.
one address · zero new UI · allowlist-first · audited
You deploy an agent, and the first question is how anyone talks to it. So you build: a chat page for humans, an API for systems, maybe a Slack bot for the team — each with its own auth, its own onboarding, its own maintenance. Meanwhile the people you built the agent for already have a perfectly good client open all day. Every request the agent could handle is one forward away, but the agent has no address to forward it to.
On a2a cloud, resources.mailbox: true provisions a real, spam-safe address on a dedicated subdomain in seconds. Email becomes the agent's front door — and the dashboard keeps the record.
Enabling the mailbox gives your agent <agent-name>@agents.a2acloud.io — a genuine, deliverable address. Anyone who can send email can reach the agent from the mail client they already live in.
All agent mail runs on the dedicated agents.a2acloud.io subdomain with its own SPF, DKIM, and DMARC. Your company domain never appears in the loop, so agent traffic can't dent its deliverability.
Out of the box, only the owner's email can write to the agent. You open the door sender by sender with allowed_senders — unknown mail never becomes an agent run.
The agent answers from its own address with proper In-Reply-To and References headers, so the response appears inside the sender's original email thread — like a colleague replying, in seconds.
Each inbound mail invokes the agent with the owner's identity and LLM credentials. Sends are locked to the agent's own From address — it can be you-shaped in capability without ever impersonating you.
Every email exchange lands in your dashboard as a chat thread with an email badge — sender messages in, agent answers out — and an audit event records each mail received and sent.
Build a chat widget, a Slack app, or an API client — one integration per audience you want to serve.
Hand out an email address. Every person and system that can send mail is already a client.
Accounts, logins, and a UI walkthrough before anyone can ask the agent a single question.
They write an email. If they're on the allowlist, the agent answers in their own thread.
Stand up a mail server or wire a third-party inbound-email service, then keep SPF, DKIM, and DMARC healthy yourself.
resources.mailbox: true in a2a.yaml — or one dashboard click — and the address is provisioned in seconds.
A public endpoint anyone can hit; you bolt on rate limits and captchas after the first incident.
Owner-only until senders are allowlisted, loop-safe filtering, and a 50-send daily cap per agent.
Mail scattered across a shared inbox; who asked what, and what the bot said, is anyone's guess.
One dashboard thread per conversation and an audit event per mail — the exchange is evidence.
Set resources.mailbox: true in a2a.yaml, or toggle it with one click in the dashboard, then deploy. Your Pro-plan agent gets <agent-name>@agents.a2acloud.io provisioned in seconds — a real address on a dedicated mail subdomain. Optionally add an allowed_senders list to say who may write to it.
Because an address is the universal interface: every stakeholder, customer, and legacy system already knows how to send email. There is zero new UI to build, host, or teach — people write to the agent from the inbox they already use, and the reply arrives in their own thread seconds later.
No. Until you allowlist senders, only the owner's email can reach the agent, and auto-responder and agent-domain mail is filtered so agents can't loop each other. An email from an unapproved sender simply never becomes a run.
It sends as itself. The agent runs with your identity and LLM credentials, but every outbound mail goes from its own <agent-name>@agents.a2acloud.io address — the SDK locks sends to that From address. Correct threading headers mean the reply still lands inside the sender's original conversation.
The mailbox is part of the agent: remove the agent and the mailbox goes with it. If your plan lapses, mail is kept for 30 days. While active, each mailbox has a 100 MiB quota and a 50-outbound-emails-per-day cap, with over-cap replies preserved in the thread and audited.
All guides live in the guides index.
a2a cloud gives every Pro-plan agent a real email address on agents.a2acloud.io — allowlist-first so only approved senders trigger runs, threaded so replies land in the sender's conversation, and audited so every mail in and out leaves an event. One line of YAML, provisioned in seconds, removed with the agent. The universal interface, governed.
