Give every deployed app an inbox

Builders ship apps constantly: preview deployments, internal tools, customer-facing agents, weekend prototypes that become production. Many of those apps need to send and receive email. Wiring a personal SMTP account into each deploy is how secrets leak and how threads get lost when the laptop sleeps. A cleaner pattern is to give every deployed app an inbox through an API call in the same provisioning path that creates the database.

Call Agent Inbox from wherever you host

Whether the app runs on Vercel, Railway, a container on your VPS, or a Neuro OS sandbox, the mail data plane can be Agent Inbox. On boot or on first user provision, create or fetch the inbox idempotently. Store the inbox id with the app’s environment config. Receive via webhook or poll. Send through the API with Ask enforced in your control plane or in Neuro OS when a role owns the work.

This is an integration pattern, not a claim about any marketplace listing. You bring the API key. You own the app. The mailbox is a dependency like object storage—environment-scoped and rotatable.

Why per-app identity matters

Preview apps should not share production mail identity. Staging should not be able to email real customers without controls. Per-deploy or per-environment inboxes keep blast radius small. When a preview is deleted, retire its mailbox. When production rotates, rotate credentials without touching developer laptops or copying .env files through chat.

Name inboxes after app and environment so support can tell which surface a customer replied to. Avoid cute shared aliases that outlive the deploy that created them.

Pair with Neuro OS when the app is an agent

If the deployed app is actually a role—support bot, onboarding concierge, ops desk—run the reasoning and connectors on Neuro OS. Keep git-defined skills, sandboxes, and Ask policy there. Use Agent Inbox as the email surface the app exposes to users and vendors. Forward escalations to humans with the thread intact so the deploy is not also the only escalation UI.

Practical checklist

Secrets only on the server. Idempotent inbox create keyed by app id and environment. Webhook signature verification. Attachment size limits. Default Ask on outbound in production. Logging that ties message ids to request ids. A destroy path that retires the mailbox when the app is torn down.

Every deployed app that must converse by email deserves an address that is not your personal inbox. Provision it like you provision the database—early, automated, and environment-scoped—so email stops being the accidental single point of human fragility.

Local development versus deployed identity

Developers often wire personal SMTP into localhost and forget to replace it. Make the app refuse to boot in staging or production without an Agent Inbox id, while allowing a clearly labeled local mock for unit tests. That single guardrail prevents the most common secret leak in agent demos: a personal password sitting in a deploy environment variable.

Keep the human path obvious

Every automation that touches external mail needs a visible escalation path. Publish who receives forwards, how Ask approvals are collected, and how to disable a role quickly. Clarity here prevents shadow processes where people quietly move work back into personal inboxes under pressure.

Agent Inbox gives each role a mailbox people can reply to, with forwarding into Neuro OS when a human must see the thread. Outbound mail defaults to Ask. Run the role on Neuro OS. To scope the first inbox, get started.

Explore Neuro OS →