How AI agents get a real inbox, pay per call instead of by the seat, and send and receive email without a human in the loop. Guides, comparisons, and deep dives from the team building DimeBox.
A practical pattern for an agent that checks its own mail, decides what matters, and acts carefully on what it finds.
Sender authentication standards exist for exactly the risk agent email has: an inbox can't tell who really sent a message without them.
If a third-party relay's spam filter drops a message before DimeBox sees it, there's no error, no bounce, no record — just silence.
Every message has a UUID. Here's how to list an inbox's full history, paginate through it, and fetch one specific message by ID.
The priority channel's guarantee is specific and worth understanding precisely, not just trusting as a marketing word.
Check the inbox yourself on a schedule, or let DimeBox push a notification the moment something arrives. Both work; they trade off differently.
If an agent feeds inbound email straight into an LLM prompt, a sender can write an email that's also an instruction. Here's what to do about it.
Nothing marks an inbound message as verified before it reaches your agent. Treat it the same way you'd treat any other untrusted user input.
Free SMTP depends on a relay's own spam filtering. Priority delivery skips that entirely, for three cents a message. Here's when each makes sense.
Plain SMTP can't complete an x402 payment mid-send, so DimeBox splits inbound mail into two separate, differently-trusted channels.
Three fields control everything about a drafted email. Here is what each one actually does, separately from the others.
Sometimes the correct answer is that no model should ever touch the outbound wording. Here is how to enforce that at the inbox level.
One request, several things happening in sequence: validation, payment, drafting, queuing. Here is each step broken out.
A send call returns before the message is actually delivered. That gap is deliberate, and it changes how an agent needs to check status.
The brief you send to the drafting endpoint is the only lever you have over what comes back. Most reliability problems trace to a weak one.
Every send costs something at DimeBox, by design. Free outbound mail from an autonomous caller is a cost someone else ends up paying.
A support agent that reads a question and drafts a reply is straightforward. Doing it safely, with content you did not write, is the actual work.
A sales agent that never forgets to follow up is mostly a scheduling problem with an email-sending step attached.
The send endpoint accepts two different request shapes. Picking the wrong one for your agent is an easy way to lose control over what goes out.
Drafting and sending don't have to be two systems glued together. One paid call against an inbox can do both at once.