An email inbox for every AI agent.
One line in a2a.yaml gives your agent a real address — <agent-name>@agents.a2acloud.io — on a dedicated mail subdomain with SPF, DKIM, and DMARC. Mail arrives, the agent runs with your identity and LLM credentials, and the reply lands back in the sender's thread in seconds. Every exchange is a transcript in your dashboard, with an audit event per mail.
# a2a.yaml+ resources:+ mailbox: true$ a2a deploy✓ mailbox provisioned (SPF · DKIM · DMARC)✓ hello-world@agents.a2acloud.io ready# inbound mail → agent runs → threaded reply← owner@example.com "summarize the attached report"→ reply from hello-world@agents.a2acloud.io — seconds later
From YAML to inbox in one deploy
Enable the mailbox
Add resources.mailbox: true to a2a.yaml — optionally an allowed_senders list — or flip it on with one click in the dashboard. Pro-plan agents only; provisioning takes seconds.
Deploy and get the address
a2a deploy provisions <agent-name>@agents.a2acloud.io on the dedicated mail subdomain. SPF, DKIM, and DMARC are already set up, so agent mail can't hurt your primary domain's deliverability.
Mail in, threaded reply out
Inbound mail invokes the agent under your identity and LLM credentials. It replies from its own address with correct In-Reply-To and References headers, so the answer lands inside the sender's original thread — in seconds, verified live.
Why an inbox beats another integration
Email is the one channel every human and every system already speaks. The mailbox turns that channel into governed agent execution — allowlisted in, capped and audited out.
Spam-safe by default
Until you allowlist senders, only the owner's email can write to the agent. Auto-responder mail and agent-domain senders are filtered, so mail loops between agents can't start. No approved sender, no run.
One conversation, one thread
Each email conversation becomes exactly one chat thread in the dashboard, threaded by the email References chain and marked with an email badge — inbound mail as user messages, agent answers as assistant messages. Full transcript, plus an audit event for every mail received and sent.
A real SDK surface, not a webhook hack
ctx.mail gives the agent list, read, send, and reply over injected IMAP/SMTP credentials, with sends locked to the agent's own From address. The @a2a.tool(on_email=True) handler contract delivers a typed InboundEmailPayload — sender, subject, body, message_id, references, attachments up to 1 MiB each.
Guardrails you can state out loud
50 outbound emails per day per agent, with over-cap replies kept in the thread and audited. A 100 MiB mailbox quota, mail retained 30 days if the plan lapses, and the mailbox removed with the agent. Bounded authority, not an open relay.
“If your agent can be reached by email, everyone your agent needs to talk to is already integrated.”
Before you deploy.
How do I give an AI agent an email inbox?
On a2a cloud it's one line in a2a.yaml — set resources.mailbox: true — or one click in the dashboard. On the next deploy the agent gets its own real address, <agent-name>@agents.a2acloud.io, provisioned in seconds. There's no mail server to run and no domain to configure.
Will agent mail hurt my domain's deliverability?
No. Every agent mailbox lives on the dedicated agents.a2acloud.io subdomain with its own SPF, DKIM, and DMARC records. Your primary domain never sends or receives agent mail, so an agent's sending behavior can't touch your company's sender reputation.
What stops strangers from spamming my agent into running?
The mailbox is allowlist-first: until you add senders to allowed_senders, only the owner's email can write to the agent. Auto-responder mail and agent-domain senders are filtered too, so two agents can't trap each other in a mail loop. Nothing triggers a run unless a sender you approved wrote it.
How does the agent actually handle an email in code?
The SDK contract is a structured handler: @a2a.tool(on_email=True) with a forced (ctx, email: InboundEmailPayload) signature carrying sender, subject, body, message_id, date, references, and attachments (inlined up to 1 MiB each). Return a string or {body, subject} and the reply is sent for you. The decorator ships in the next a2a-pack release; the inbox, threads, and replies are live today.
What are the limits on an agent mailbox?
Each agent can send 50 outbound emails per day — over-cap replies stay in the dashboard thread and are audited, they just don't go out over SMTP. The mailbox holds 100 MiB, mail is kept 30 days if the plan lapses, and the mailbox is removed with the agent. Every received and sent mail also writes an audit event.
Trust the thread it can't edit.
a2a cloud gives every Pro-plan agent a real inbox on agents.a2acloud.io — allowlist-first, capped at 50 sends a day, with every received and sent mail recorded as an audit event and every conversation preserved as a dashboard transcript. Enable it with one line of YAML, deploy, and email your agent.