About
Email infrastructure for agents somebody else built
Operated by Genetech Software Solutions, based in California, United States.
Email for Agents does one thing: it gives an AI agent an email address, the tools to use it,
and a human a way to watch and stop it. We do not build agents, host agents, or have an
opinion about which framework you should use. Your agent stays in the application you already
run; we supply the inbox.
Why this exists
Because the two obvious options are both bad.
Handing an agent your mailbox
An OAuth token on a personal Gmail or Microsoft 365 account gives an agent your identity,
your entire history and your contact list. Every message it sends is from you. Scoping
that down is awkward, and revoking it disrupts a human's own mail.
Building it yourself
Inbound routing, MIME parsing, attachment scanning, threading, idempotent sends, delivery
events, webhook signing and retries, approval state. It is a quarter of work before the
agent does anything interesting, and it all has to be right.
How we decide what is true here
A marketing site is easy to write and hard to keep honest, so the honesty is enforced rather than intended.
-
Limits come from the code. Every number quoted on this site — 3 test
inboxes, 100 retained messages, 10 MB, 7-day content expiry, the 30-day idempotency
window, the webhook retry schedule — is read out of the implementation, not chosen for
the page.
-
A test blocks the claims we are not allowed to make. An integration test
scans every source file on this site and fails the build on an uptime percentage, a
compliance badge, a delivery guarantee, a capacity claim with no cap behind it, a
certified client, or a testimonial.
-
Core mail does not call a language model. Permissions, approvals,
policy and webhooks are ordinary application code and SQL. The agent is the model; the
mail path is not.
What the first release deliberately leaves out
Shipping a short list of finished things beats shipping a long list of nearly-working ones. These are the gaps, stated once, up front.
No published SDK packages
There is no TypeScript or Python package to install. The REST API is stable and documented, and every sample on this site is plain HTTP.
No paid plans yet
Stripe checkout is disabled. Builder and Team are proposed plans at proposed prices; they are not available to buy, and final terms will appear before any charge.
No custom domain yet
Domain verification is built and documented, but it is a paid-plan feature — and there is no way to reach a paid plan today. Live mail runs on agents.emailforagents.ai.
No certified client
No MCP client is certified, supported or one-click. Every OAuth connection is recorded with a trust class of unverified, whichever application opened it.
No compliance certification
No SOC 2, ISO, HIPAA or GDPR badge, and no independent security review has been commissioned for launch. The security page lists what the code enforces instead.
No uptime promise
There is no published service-level agreement, no percentage, and no third-party status board. The status page lists the surfaces that exist and where to check them.
No full-text search
Search covers stored subjects, senders and previews. Complete message-body and attachment-content search are not included.
No claim to be prompt-injection proof
A message cannot widen a grant — scopes are re-derived on every request — but an agent can still be argued into a bad draft. That is why the approval queue exists.
What changes, and when, is written down in the changelog.