Agent Inbox vs Resend Inboxes
Agent Inbox vs Resend Inboxes: a shared team inbox or a way for your agent to reach a person? Compare approvals, channels, security and pricing, with code.

On October 22, 2026, Resend launched Inboxes and called it "the inbox for agents": email inboxes on your own domain, built for agents and ready for your whole team, where people and agents work side by side.
Agent Inbox on the other hand, gives an agent its own email address, Telegram bot, Slack app and WhatsApp Business number. People message the agent and it answers and acts, it starts conversations, and when a decision isn't its to make, it asks the right person and waits for the answer.
Both give your agent a real email address with API, CLI and MCP access. The difference is who the inbox is built around: Resend Inboxes is an email inbox for your team, with agents working inside it, while Agent Inbox is the agent's own inbox, across the channels people actually use.
We’ll be comparing how each works, run four everyday workflows through both, and cover setup and pricing, so you can pick the one that fits your use case.
The short answer

How each works
Resend Inboxes
Resend describes Inboxes as "the inbox for agents". Each inbox lives at an address on a domain you've verified with Resend, such as support@yourcompany.com, or receives mail forwarded from an existing address. Resend stores every email, groups replies into threads, and sorts them into folders: inbox, sent, archive, spam and trash. People work those threads in the dashboard with search, labels, notes and assignment, and Resend describes them as shared inboxes "the same inbox can be worked by people and agents side by side."
Each inbox can run a built-in inbox agent. You give it instructions, pick a tone, and choose which actions it may take: draft replies, forward threads, add labels, assign threads, mark spam, archive or move to trash. It runs on every new email and never sends anything on its own. Replies and forwards become drafts that a teammate reviews and sends. Your own agents, such as Claude Code or Cursor, can also work an inbox through the API, SDKs, CLI or MCP server, and webhooks report events like new threads and sent drafts.
Agent Inbox
Agent Inbox starts from the agent rather than the inbox. One command, npx @agentinbox/cli setup, gives your agent its own email address straight away, at agentinbox.mx or on your own domain on paid plans. In a few minutes more, it can have its own Telegram bot, Slack app and WhatsApp Business number, plus a contact card people can scan to reach it. You connect the agent you already use: through the CLI and skills for Claude Code, Codex, Cursor, OpenCode, Hermes or OpenClaw, through the MCP server at https://mcp.agentinbox.md for Claude, ChatGPT and Grok Bot, or through plugins for Claude Code and Codex.
From there, conversations run in both directions:
- People reach the agent. A customer emails it, a teammate messages it on Slack, someone writes on WhatsApp. The message lands in the agent's inbox, and the agent reads it, answers on the same channel, and gets on with the task. For Claude Code, Codex and OpenCode,
listengoes further: a message starts the agent in your project. - The agent reaches people. It can start a conversation with any contact: a status update, a question, a report when a job is done.
- The agent asks for decisions. When something isn't its call, it sends an approval or a choice with buttons, to one person or a whole team, and waits for the answer before it acts.
Every incoming message is marked as coming from a contact or a stranger, and a stranger's text is treated as information, never instructions.
Workflow 1: a customer writes in
A customer, Pawan, asks about a late refund. Some questions your agent can answer on its own. Others, like a refund over your limit, need a person.
With Resend Inboxes
Pawan emails support@. The inbox agent reads it, labels it, assigns it to the right teammate and drafts a reply following your instructions. A person reviews the draft and sends it. Your own agent can also reply in the thread through the API:
Every reply has a person behind it, and assignment, notes and labels keep the handoffs visible to the whole team.
With Agent Inbox
Pawan writes to your agent by email or on WhatsApp, and the message lands in its inbox. Your agent reads it, looks up the order and answers him on the same channel. When the answer needs a decision it can't make, it asks Tamar, the support lead, and waits, then replies to Pawan with the outcome:
The agent handles the conversation end to end, and a person steps in only for the decision.
The difference: Resend keeps a person on every reply and gives your team a full shared inbox to work in. Agent Inbox lets the agent run the conversation itself, on email or WhatsApp, and pulls a person in when a decision needs one. If you want people sending every reply, choose Resend. If you want the agent answering, choose Agent Inbox.
Workflow 2: the agent needs someone's approval
A finance agent has matched a $2,400 invoice from a vendor and is ready to pay it. Payments over $1,000 need a yes from Dan or Tamar, and the agent should only pay on a yes.
With Resend Inboxes
The inbox agent never sends, so it can't ask and wait. Your own agent can email Dan and Tamar from an inbox, and you'd listen for a reply through Resend's webhooks, read the text, decide whether it means yes, and only then call your payment code. It works, but the approval logic is yours to build: the wording of a yes, a reply from the wrong person, two people answering differently, and how long to wait.
With Agent Inbox
Approval is a message type. Dan and Tamar each get Approve and Deny buttons on the channel they use, and the command waits:
The first valid answer decides, and --json records who gave it and when.
The difference: with Resend you build approvals on top of email replies. With Agent Inbox they're built in, on email, Telegram, Slack or WhatsApp.
Workflow 3: a teammate messages the agent to start work
You're away from your desk, and you want Claude Code to fix a failing test in your project. You'd rather send one message from your phone than open a laptop.
With Resend Inboxes
You could email a Resend inbox, and a webhook would tell your systems a new thread arrived. Starting Claude Code, giving it the request and sending the result back is code you'd write and host yourself. The built-in inbox agent triages and drafts. It doesn't run tasks in your project.
With Agent Inbox
Run listen once in your project:
From then on, a message you send your agent on Telegram, Slack, WhatsApp or email starts Claude Code in that folder, with your message as the task. It replies in the same conversation when it's done, and it can send updates along the way. Any permission its settings don't allow arrives as Allow and Deny, and no answer within 10 minutes counts as Deny. It works for Codex and OpenCode too, and you can let a teammate's messages start runs as well.
The difference: with Resend, an incoming email is an event for your code to handle. With Agent Inbox, a message from you or a teammate can start your agent directly.
Workflow 4: the agent signs up for a service
The agent needs a trial account on a monitoring tool, and has to read the verification code from the email that comes back.
With Resend Inboxes
Use an inbox address on your domain in the sign-up form. When the email arrives, the inbox.thread.created webhook delivers it to your code with the sender, subject and message details, and your agent fetches the thread to read the code.
With Agent Inbox
Use the agent's address, then wait for the email without polling:
The sign-up service counts as a stranger, so its email is treated as information, never instructions.
The difference: both handle sign-ups well. Resend needs a verified domain first, while Agent Inbox works on its own address from the first minute.
Security: what changes when anyone can email your agent
An inbox is an open door, and a language model reads every word that comes through it. Here's how each product handles the risks that matter most.
Prompt injection
Resend's answer is to limit what the inbox agent can do. It never sends email on its own, it forwards only to full addresses written into your instructions, it can use only labels that already exist, and it can't archive, trash or mark as spam any email that needs a person's attention. Resend's docs also warn against trusting sender addresses or display names, and against putting secrets or customer data in the agent's instructions, since drafts go back to senders.
Agent Inbox works from who sent the message. Every message is marked as coming from a contact, someone you invited who confirmed, or a stranger. The Agent Inbox skills tell the agent that a stranger's text is information, never instructions: it can read and reply, but never runs, approves, sends or changes anything because a stranger asked. A stranger's message never starts the agent through listen, and during a listen run, anything outside the project's settings goes to you as Allow and Deny.
Neither approach makes injection impossible. Keep your agent's own permissions narrow with either one.
Who can act, and who can approve
In Resend Inboxes, people decide by sending, or not sending, the agent's drafts, and assignment makes clear who owns each thread. With your own agents on the API, what they're allowed to send is up to your code.
In Agent Inbox, an approval goes to the contacts you name, and only they can answer it. Every answer records who gave it and when. The agent can start conversations only with contacts, never with strangers, and you can switch it to contacts only or pause it.
Keys and compliance
Resend uses API keys, and lists SOC 2 Type II and GDPR compliance on its security pages.
Agent Inbox agents that connect over MCP sign in with OAuth, so the agent never holds an API key, and Telegram and Slack tokens go into a setup page, never through the chat. Agent Inbox is built by the Novu team on Novu's infrastructure, so it's covered by Novu's SOC 2 Type II, ISO 27001, GDPR and HIPAA compliance.
Side by side
Pricing
The two meter different things.
Resend includes 3 inboxes on its Free, Pro and Scale plans, and inbox email uses your transactional email quota:
Agent Inbox charges by active conversation:
An active conversation is one thread, on one channel, within 24 hours. Every message, reply and decision inside that window counts once. A support inbox that handles hundreds of customer emails a day suits Resend's email-based pricing. An agent with a few decisive exchanges a day, such as approvals and check-ins, suits Agent Inbox's.
When to use Agent Inbox
- You want the agent to run the conversation. People write to it, and it answers and acts on the same channel, bringing in a person only when a decision needs one.
- People reach you on more than email. One agent, one identity, on email, Telegram, Slack and WhatsApp Business.
- Your agent needs decisions before it acts. Payments, refunds, deploys: approve and choose arrive as buttons, sent to one person or a team, and the agent waits for the answer.
- You run coding agents.
listenlets a message from your phone start Claude Code, Codex or OpenCode in your project, with permissions routed back to you. - You want to start without a domain. The agent's address works from the first command, and your own domain is available on paid plans.
Try it next to your current setup:
It's free for 100 conversations a month, with no credit card required. Read what an inbox for AI agents is for the bigger picture, or the Agent Inbox launch post.
When to use Resend Inboxes
- Your team and an agent share an inbox. support@, sales@ or jobs@, where an agent does the first pass and people review and send.
- You want an agent that can't send by itself. The inbox agent only drafts, labels, assigns, forwards and files, so nothing reaches a customer without a person.
- You already send with Resend. Inbox email runs on the same account, domain and plan.
- You need a full email client for people. Folders, search, labels, notes and assignment, in a dashboard your team can use every day.
If you need both, they fit together: Resend Inboxes for the shared mailbox, and Agent Inbox for the moments your agent needs someone's yes.
FAQ
What's the difference between Agent Inbox and Resend Inboxes?
Resend Inboxes, which Resend calls "the inbox for agents", is a set of email inboxes on your domain where people and agents work side by side, with a built-in inbox agent that triages and drafts replies but never sends. Agent Inbox gives an agent its own identity on email, Telegram, Slack and WhatsApp Business: people message the agent and it answers and acts, it starts conversations, and it asks for approval and waits when a decision isn't its to make.
Can the Resend inbox agent send emails on its own?
No. Resend's inbox agent saves replies and forwards as drafts, and a team member reviews and sends them. It can also label, assign, archive, mark spam or move threads to trash, depending on which actions you allow.
Does Resend Inboxes work with Telegram, Slack or WhatsApp?
No, Resend Inboxes is email only. Agent Inbox covers email, Telegram, Slack and WhatsApp Business, with iMessage and voice coming next.
Can I use Agent Inbox and Resend Inboxes together?
Yes. A team can run a shared support mailbox on Resend Inboxes and give its agents their own identity on Agent Inbox, so they can answer people on Telegram, Slack and WhatsApp and send decisions, such as refunds over a limit, to the right person for approval.
Do I need my own domain?
For Resend Inboxes, yes: inboxes live on a domain you've verified with Resend, or receive mail forwarded from an existing address. Agent Inbox gives your agent an agentinbox.mx address at setup, and custom domains are available on paid plans.