BYO email for people and agents

One inbox for all your accounts, shared with your agents.

Connect the mailboxes you already have, as many as you like, and read them together in one app that runs on your own machine. Your agents use the same accounts through the CLI or MCP, and whatever they draft waits for you.

Install on macOS or Linux

curl -fsSL https://u1f4e7.com/install.sh | bash
brew install tymrtn/u1f4e7/envelope

The script checks the release's SHA-256 and installs to ~/.local/bin without sudo. Homebrew builds from source, so it installs Rust first.

Add it to your agent

/plugin marketplace add tymrtn/U1F4E7
codex plugin marketplace add tymrtn/U1F4E7

Then install envelope@envelope. The plugin adds a setup skill that walks your agent through connecting your accounts; you create the agent token yourself. Cursor: install Envelope from the Cursor Marketplace.

Current release: Envelope 1.3.16. Since 1.3.0, rShield gives every incoming message a local phishing and malware score from six analyzers.

  • Every account in one Unified Inbox
  • Search all your accounts at once
  • Bring your own mailbox
  • CLI + JSON + MCP for Claude Code, Codex, OpenHands, Hermes
  • Agent drafts wait for your approval

Use cases

One runtime. Several buying moments.

Many accounts, one inbox

Gmail, Fastmail, iCloud and your own domain land in one Unified Inbox, newest first. Search every account at once with Gmail-style operators, and archive or snooze across accounts in one pass. Each account keeps its own server, folders and address.

See the unified inbox →

Shared inbox agents

Add an agent to support@, ops@, editor@, or your personal inbox. Same mailbox, same workflows, better coverage.

Read the shared inbox page →

Agent questions: Can an AI agent safely read 2FA codes from your email? and How do I make an AI agent ask before it sends email?

Named workflows

The things agents actually need from email.

OTP Relay

Your agent calls envelope code --wait 60 and blocks. When the OTP arrives, the CLI returns it. No polling loop, no parser, no race condition against your own inbox.

Reviewable Rules

Agent proposes rules based on what it found in your inbox. You see exactly which messages each rule would touch before anything moves. Confirm, edit, or reject. Rules export to Sieve if your server supports it.

Local dashboard

A mail client and an agent control plane on localhost. Read, search, star, archive, and triage mail, with keyboard shortcuts on the message list (c compose, x select, e archive, s star). The Review page collects drafts waiting for approval, scheduled sends, and failed auth; the Logs page keeps the history of what each agent did. Approve, edit, or discard agent work from one place.

Evidence Bundle

Export a thread as raw RFC822 files with a manifest and checksums. Useful when you need a verifiable record — for legal holds, compliance review, or a paper trail your agent assembled during an investigation.

Human control

The agent doesn't send anything you haven't approved.

By default, every draft goes into a review queue. Rule runs preview affected messages before anything moves. The Logs page shows the full history. You can pause, override, or stop any workflow.

This isn't a safety flag bolted on at the end. Agent contexts default to draft-only mode. Autonomous send is an option, but you have to opt in explicitly — and every action is logged either way.

Send modes

  • draft-only — agent composes, you send (default)
  • confirm-send — agent sends after you approve
  • allowlisted-send — autonomous send to approved recipients
  • autonomous-send — explicit opt-in, everything logged

Every send decision produces a policy audit event. No secrets in the logs.

Works with your existing stack

Any harness. Most IMAP mailboxes.

If it can run a CLI command or speak MCP, it works with Envelope. Mailboxes that sign in with a password or app password work today, including Gmail, Fastmail, iCloud Mail, and Migadu; Outlook.com and Microsoft 365 need OAuth sign-in, which Envelope does not support yet. The same JSON surface works from a shell script, a Hermes pipeline, a Claude Code tool call, or a Codex exec loop.

Claude Code

Drop in via MCP. Full inbox access as a tool call.

OpenAI Codex

CLI + JSON surface for terminal-native autonomous loops.

Hermes

First-class skill for Hermes profiles and multi-agent pipelines.

OCPlatform

OTP handling, watch events, and thread-aware follow-up.

OpenHands

Long-running loops that wait on email before continuing.

Tailscale

Secure, identity-aware dashboard access across devices with Tailscale Serve.

Why not the alternatives?

Hosted services ask you to change. Envelope doesn't.

Robotomail, Cloudflare Agentic Inbox, and similar

These provision a new mailbox on their infrastructure. New address. DNS changes. Your mail in someone else's system. If your agent needs to operate on the inbox your contacts already use, you're at the wrong starting point.

Envelope connects to the mailbox you already have. Nothing migrates.

Mailgun, SendGrid, Postmark, Resend

If all you need is authenticated SMTP and a small API for product email, paying a third-party relay can be pure duplicate infrastructure. Envelope gives your own server a simple mail runtime without surrendering the mailbox or adding another vendor bill.

For huge marketing sends, use a bulk sender. For ordinary app email and agent workflows, BYO email is saner.

Himalaya and other IMAP CLIs

Solid tools built for humans reading their own mail. Envelope is built for the case where an agent is the primary consumer: IMAP IDLE watch, OTP extraction, JSON everywhere, reviewable rules, and a control plane for the human sharing the inbox with the agent.

Annual licensing

Personal use is free. Commercial rollout is a flat annual license.

Annual bands sized by mailbox users on your domain. No per-message metering, no seat trickery. Every paid band also lifts the free tier's two-agent cap.

Personal

Free

For individuals running their own mailbox

  • Personal, non-commercial use on your own mailboxes
  • Full CLI, dashboard, and MCP surface
  • Up to 2 active agent identities (a third asks for a license key)
  • Community support via GitHub issues
Get the repo

Team

$240 /year

Up to 10 mailbox users on one domain

  • Commercial use license for one organization
  • Up to 10 mailbox users across one primary domain
  • Unlimited agent identities
  • Email support, best-effort next-business-day response
  • Annual renewal, invoiced
Request annual license

Growth

$960 /year

Up to 25 mailbox users, one domain

  • Commercial use license for one organization
  • Up to 25 mailbox users across one primary domain
  • Unlimited agent identities
  • Priority email support, next-business-day response target
  • Annual renewal, invoiced
Request annual license

Enterprise

Contact us

26+ users, multi-domain, or embedded/OEM

  • Multi-domain or large-team deployments
  • Embedded/OEM and reseller terms available
  • Custom support agreement and security review
Talk to sales

Licenses are billed annually. Quotes and invoices on request; payment by card or bank transfer. Email ty@tmrtn.com for anything not covered above.