Early access

Your AI agent is about to send that email. Are you sure?

The approval layer between AI agents and your inbox.

Email has no git. Until now.

Nuvrail sits between your AI agent and your inbox. Every proposed action is staged for human review, logged immutably, and held until you approve the diff.

From the blog

Email is the Most Dangerous Thing You Can Give an AI Agent

2026-05-19

It's 7:43am. Your AI assistant has been running overnight, triaging your inbox. You get to your desk and find that the proposal you were still editing — marked as a draft, saved but not sent — went out at 3am. To your biggest prospect. Without the pricing section. With a placeholder still in the subject line that says “SUBJECT TBC.”

These aren't hypotheticals. Variations of all three have already happened. They will keep happening as long as we keep giving AI agents raw IMAP access and hoping for the best.

Read the full post

The problem

AI agents that touch email today have unrestricted access — no review mechanism, no audit trail, no undo. When something goes wrong, there's no record of what happened and no way to reverse it.

That's not a risk you can accept.

The solution

Nuvrail is the pull request model for email.

Every proposed action — mark as read, move to folder, send a reply — is staged for your review before anything touches your real inbox.

You see a readable diff. You approve or reject. The agent proceeds only when you say so.

app.nuvrail.com
Nuvrail1 pending
SMTPSend email to vendor@newservice.com

To: vendor@newservice.com

Subject: “Let’s sign this contract”

Queued 2 minutes ago·Expires in 47 hours

How it works

01

Connect

Your AI agent connects to Nuvrail using standard IMAP/SMTP credentials. No code changes required on the agent side.

02

Stage

Every proposed action is staged — you see exactly what the agent wants to do, presented as a readable diff before anything happens.

03

Decide

You approve or reject. Approved actions execute against your real inbox. Rejected actions are cleanly reversed. The agent only proceeds when you say so.

Built for trust

Open source core

The proxy, staging logic, and audit schema are public. Read exactly how it works.

Immutable audit log

Every proposed, approved, and rejected action is recorded forever. Nothing disappears from the record.

No permanent deletes

The gateway blocks EXPUNGE. Nothing is gone until you explicitly say so — with a record of exactly when and why.

We never read your email

Nuvrail is a proxy, not a reader. We see action metadata — not message content. Your email stays yours.

Get early access

Be first when we open the doors.

We're building with a small group of early-access teams. If you're deploying AI agents on email infrastructure, we want to hear from you.

Early access + architecture notes from the team.

No spam. We'll reach out when we're ready for early access teams.