self-hosted agent email vs managed: what actually makes sense

Your agent needs email. You've settled on that much. Now you're staring at two paths: run the infrastructure yourself, or pay someone else to run it. Both paths work. Neither is free of consequences.

The self-hosted camp includes projects like AgenticMail (open-source, purpose-built for agents), Postal (general-purpose mail server), Mail-in-a-Box (full stack on one box), and raw AWS SES with your own application layer on top. The managed camp includes LobsterMail, AgentMail, and to a lesser extent the traditional transactional providers like SendGrid and Resend.

This isn't a "managed is better" post. Self-hosting is a legitimate choice for certain teams. But the decision involves more variables than most comparison posts acknowledge. Here's what actually matters.

What self-hosting gets you#

Full control. That's the honest answer, and it's not a small thing.

When you run your own mail server, you own the entire stack. You decide which MTA handles routing. You control the queue depth, the retry logic, the rate limits. If a managed provider changes their terms, raises prices, or goes down, you're unaffected. Your data never touches someone else's infrastructure.

AgenticMail specifically targets the agent use case. It's open-source, so you can inspect every line of code that handles your agent's messages. If you need to modify how inbound emails get parsed or add custom preprocessing before your agent sees the content, you fork the repo and build it.

For teams with strong DevOps skills and strict data residency requirements (GDPR compliance in a specific EU region, government contracts with FedRAMP constraints), self-hosting might not just be preferable. It might be mandatory.

What self-hosting costs you#

Here's where the comparison posts usually hand-wave. "Just spin up a VPS" sounds simple until you list everything that VPS needs to do.

DNS configuration#

Email requires correct MX records, SPF TXT records, DKIM key pairs, and DMARC policies. Get any of them wrong and your messages silently fail or land in spam. SPF has a 10 DNS lookup limit that's easy to exceed. DKIM keys need to be 2048-bit minimum. DMARC needs to start at p=none and ramp to p=reject over weeks of monitoring. This isn't hard for someone who's done it before. It's a minefield for someone who hasn't.

IP warming#

A fresh IP address has zero reputation. Send 500 messages on day one and Gmail will throttle or reject them. The standard warmup process takes four to eight weeks of gradually increasing volume, starting with your most engaged recipients. During warmup, your agent's emails will have inconsistent delivery. Some will arrive. Some won't. You'll need to monitor bounce rates, complaint ratios, and inbox placement daily.

If you're running AgenticMail on a $10/month VPS with a single IP, that IP is your entire reputation. One misconfiguration, one spam complaint spike, and you're requesting delisting from Spamhaus while your agent sits silent.

Ongoing maintenance#

Mail servers don't run themselves. OpenSSL patches. TLS certificate renewals. Storage management as mailboxes grow. Queue monitoring when downstream servers reject your messages. Log analysis to catch delivery problems before your users notice. OS-level security updates that can break your MTA configuration if you're not careful.

Postal, for example, is solid software. But its GitHub issues tell the story: users troubleshooting DKIM failures after upgrades, dealing with RabbitMQ memory leaks, and debugging MySQL lock contention under load. These aren't Postal bugs. They're the operational reality of running a mail server.

Deliverability management#

This is the part that surprises people. Sending email is easy. Getting email delivered to the inbox instead of the spam folder is an ongoing discipline.

Google requires bulk senders to maintain a spam complaint rate below 0.3%. Microsoft uses a reputation system that weighs domain age, sending patterns, and recipient engagement. Both change their filtering rules without notice. When your agent's emails stop arriving, you need to diagnose whether it's an SPF alignment issue, a content fingerprinting problem, or a reputation dip from a burst of sends your agent triggered at 3 AM.

Managed services have teams that handle this full-time. When you self-host, you're that team.

Security and compliance#

Every internet-facing mail server is a target. SMTP is one of the oldest and most attacked protocols on the internet. Open relays, brute-force authentication attempts, and TLS downgrade attacks are daily events. You'll want fail2ban or equivalent, regular penetration testing, and a process for responding to abuse reports that ISPs send when something goes wrong.

If your agent handles sensitive data through email, you're also responsible for encryption at rest, access controls, audit logging, and potentially compliance certifications that your self-hosted setup needs to satisfy independently.

What managed services get you#

Speed and operational leverage. Your agent has a working inbox within minutes, not weeks. DNS records are either pre-configured (on the provider's domain) or generated for you during custom domain setup. IP reputation comes from a shared or pre-warmed pool. Deliverability is monitored by people whose full-time job is keeping emails out of spam folders.

LobsterMail, for example, runs on AWS SES with per-account configuration sets for reputation isolation. DKIM, SPF, and DMARC are handled during domain setup. The infrastructure underneath (Lambda, ECS Fargate, RDS PostgreSQL, ElastiCache, S3) is managed and patched without you thinking about it. Your agent calls one function and starts communicating.

The tradeoff is dependency. You're trusting a third party with your email pipeline. If they have an outage, your agent's email goes down. If they change pricing, you adapt or migrate. If they get acquired or shut down, you need a new plan.

The honest comparison#

When self-hosting makes sense#

You have a dedicated DevOps team with email infrastructure experience. Your compliance requirements mandate specific data residency or air-gapped deployment. You need deep customization of the mail pipeline that no API can provide. You're sending at volumes where managed pricing becomes prohibitive and you've already solved deliverability.

If three or more of those apply to you, self-hosting is worth the investment.

When managed makes sense#

You want your agent sending email this week, not next month. Your team's time is better spent on what the agent does rather than how it delivers messages. You don't have email infrastructure expertise in-house, and you'd rather not develop it. You're running fewer than 100,000 emails per month and the managed pricing pencils out.

For most teams building AI agents today, this is the reality. The agent is the product. The email infrastructure is plumbing. Nobody gets a competitive advantage from running their own SMTP server.

The middle ground#

Some teams start managed and migrate components to self-hosted as they scale. Use a managed service for the first year while you're iterating on your agent's capabilities and volume is low. If you hit a point where the managed approach creates real constraints, you'll have the operational knowledge and the traffic patterns to make self-hosting work.

Others go hybrid: self-hosted inbound processing with managed outbound delivery through SES or a similar provider. This gives you control over how your agent consumes incoming email while offloading the hardest part (deliverability) to infrastructure built for it.

There's no single correct architecture. But there is a wrong time to self-host: when you're still figuring out whether your agent even needs email in the first place. Solve the product problem first. Optimize the infrastructure later.

---

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