← All posts

Can an AI agent have its own phone number and email? From Microsoft's agent identities to Manus Cue

Can an AI agent have its own phone number and email? Vestra blog banner

Yes. An AI agent can have its own phone number and its own email address today, and some makers now give agents a full identity of their own, with a name, an inbox, a calendar and a person who answers for them. The real question for a business is not whether an agent can have a number, but who that agent belongs to, who signs off on what it says, and whether your agents work as one team or as strangers with separate inboxes.

This guide explains what an agent identity is in plain words, shows exactly who gives agents a number, an inbox or an identity today (Microsoft, Google, Manus Cue, OpenAI, Lindy, Tasklet, xAI and others), and gives you simple rules for setting one up safely. Every competitor fact comes from the maker's own pages, checked on 23 November 2026. Products in this space change every few weeks, so check the maker's page before you rely on a detail.

Quick answer

Can an AI agent have its own phone number and email? Yes. Here is the short version.

  • Phone numbers: Manus Cue, Lindy and Bland's Agent Phone Plan give an agent a number of its own. Tasklet gives an agent a texting number. Voice platforms like Vapi, Retell and Twilio let developers buy a number and connect it to an agent.
  • Email addresses: Microsoft Copilot Autopilot, Microsoft Foundry agents, Google's Gemini agent for work, Manus Cue and Tasklet give an agent its own address. AgentMail sells inboxes for agents through an API.
  • A full identity: Microsoft gives each Autopilot its own Microsoft Entra Agent ID and a named manager. Google gives its coworker agents their own Workspace account. Okta and Auth0 sell identity tools for agents.
  • Who it belongs to: most of these agents belong to one person (Autopilot, Cue, dots, Lindy) or to IT (Foundry, Agent 365). Very few are built so that a whole team of agents works together under one set of rules.
  • Vestra today: your chief has its own phone number and its own email address now, with phone calls from the Pro plan up. A phone and an inbox for every teammate is on Vestra's roadmap, not live yet.

Key takeaways

  • An agent with its own number and inbox is easier to trust, because everything it sends is clearly its own and not yours.
  • Identity is more than a number. It is a name, a login, a manager, a set of permissions and a record of what the agent did.
  • Microsoft and Google now treat agents almost like employees, with accounts in the company directory. That is powerful, but it lives inside their stack and is set up by IT.
  • Manus Cue gives one person a set of personal agents, each with its own phone, email, wallet and computer. It is built for personal life, and its terms say it is for personal purposes.
  • Most other tools give one agent one channel: a texting number here, a shared email address there.
  • The hard part is not giving an agent a number. It is making sure every agent with a number follows the same rules: first contact waits for a person, money waits for a person, and every call and email lands on one record.
  • Before your agent calls or texts anyone, learn the basic rules: US law treats AI voices in robocalls as artificial voices, US business texting needs registration, and Gmail and Yahoo require email authentication from bulk senders.
  • Five personal agents with five numbers are still five strangers. A team needs one place where the agents share memory, rules and receipts.

Contents

  • Can an AI agent have its own phone number and email?
  • What is an AI agent identity?
  • Why would an AI agent need its own phone number and email?
  • Who gives an AI agent its own number, inbox or identity today?
  • How does Microsoft give AI agents their own identity?
  • How does Google give AI agents their own identity?
  • What is Manus Cue, and what does each agent get?
  • Do OpenAI dots have their own phone number?
  • What do Lindy, Tasklet, Grok Bot, Devin Bots, Viktor, Hermes and Claude offer?
  • Can you build your own agent phone number and inbox?
  • Who does an AI agent with its own identity belong to?
  • What goes wrong when every agent has its own identity but no shared rules?
  • How does Vestra handle an agent's phone number and email?
  • What rules should an AI agent with a phone and email follow?
  • How do you set up an AI agent's phone number and email safely?
  • How do you take an agent's phone number and email away?
  • What does a week look like with an agent that has its own number?
  • How much does an AI agent phone number and email cost?
  • Which option fits which business?
  • Frequently asked questions
  • Glossary
  • Sources

Can an AI agent have its own phone number and email?

Yes. In November 2026 several products give an AI agent a phone number, an email address or both, and a few give it a full identity in a company directory. What differs is whether the agent can call or only text, whether the address is its own or shared, and who controls it: one person, IT or a team.

A year ago, most AI agents borrowed your accounts. They sent email from your Gmail, drafted messages for you to send, or had no way to reach anyone outside a chat window. That is changing fast. In the last few months, Microsoft, Google and Manus each announced agents that hold accounts of their own.

Here is the plain state of things.

  • An agent can have its own email address. Microsoft says each Copilot Autopilot has "its own mailbox and calendar alongside yours". Google says its Gemini agent for work "receives its own Workspace account, including an email address, calendar, Drive, and presence in your company directory". Manus Cue says an agent can "send and receive email from its own address". Tasklet gives each agent an address like research-assistant@tasklet.bot.
  • An agent can have its own phone number. Manus Cue says an agent can "make calls, take calls, and send messages from its own number". Lindy's docs say "Lindy can have its own phone number", with a US or Canadian area code, on a paid plan. Bland sells "a real US local number that belongs to the AI you already use".
  • An agent can have its own identity. Microsoft says "Each Autopilot has its own Microsoft Entra Agent ID and a named manager." Google says "Every agent gets its own identity, cryptographically attested and governed like an employee."

So the technology is here. The questions that matter for a small business are different ones:

  1. Who does the agent work for, and who answers for what it says?
  2. Can it contact people you have never dealt with, on its own?
  3. Can it spend money, and with what limit?
  4. If you run several agents, do they know about each other, or does each one only know its own inbox?
  5. When you stop using an agent, how do you take its number and inbox back?

The rest of this guide answers those questions, tool by tool.

What is an AI agent identity?

An AI agent identity is everything that marks an agent as a distinct actor: its name, its own login or account, its email address and phone number if it has them, the permissions it holds, the person who answers for it, and a record of what it has done. It tells the agent's actions apart from yours.

People often use "identity" to mean only a login. For a business, it helps to split it into six parts. Each one can exist without the others.

The six parts of an agent identity

  1. A name. What people see when the agent writes or calls. Microsoft says Autopilot's "name, appearance, email address, and Microsoft Entra identity distinguish its contributions from yours."
  2. An account or credential. The thing a computer system checks before it lets the agent in. In Microsoft's world that is a Microsoft Entra Agent ID. In Google Cloud it is called Agent Identity, which Google describes as "a strongly attested, cryptographic identity for each agent that is based on the SPIFFE standard."
  3. Contact points. An email address, a phone number, a texting number or a chat handle that belongs to the agent and not to you.
  4. Permissions. What the agent may open, read, change, send or spend. A good identity carries only the permissions the job needs.
  5. A responsible person. Someone who answers for the agent. Microsoft calls this the agent's manager. Okta's page says it lists agents in one directory with a human owner for each.
  6. A record. A log of what the agent did, so you can check it later. Without a record, an identity is just a name.

Identity versus impersonation

There is an important line here. An agent with its own identity speaks as itself. An agent using your accounts speaks as you. The second is quicker to set up, but it blurs who did what. If an agent sends an email from your Gmail, the person reading it thinks you wrote it. If it sends from its own address, the reader can see it came from your agent.

Most people want the first. It is more honest with the people you deal with, and it is safer for you, because a mistake by the agent is visibly the agent's.

Short definitions you will meet in this guide

  • Agent ID: a credential that identifies one agent to a computer system, like a staff badge.
  • Mailbox: a store of email that belongs to one account, with an address people can write to.
  • Phone number: a number the agent can call from, take calls on, or text from. Some numbers do only one of these.
  • Wallet or vault: a store of payment details and logins the agent may use without seeing the raw numbers or passwords.
  • Manager or owner: the person who answers for the agent and sets its limits.

Why would an AI agent need its own phone number and email?

Because work involves other people. An agent that books suppliers, chases invoices, follows up with leads or calls a clinic needs a way to reach them. A number and inbox of its own keeps your personal accounts separate, makes clear who sent what, lets you take access back cleanly, and gives every message a place on the record.

You can run an agent through your own accounts. Many people do. But it causes small problems that add up.

Five reasons to give an agent its own contact points

1. It keeps your own accounts yours. If an agent works from your inbox, it can read everything in it, not just what the job needs. A separate inbox holds only the agent's work. Your personal email, your bank alerts and your private threads stay out of reach.

2. It makes clear who said what. When someone gets a message from the agent's own address, they know it came from the agent. If something goes wrong, you can find out which agent sent it and why. When everything comes from your address, nobody can tell, including you a month later.

3. It lets you take access back cleanly. If an agent uses your Gmail login and you want to stop it, you have to change settings on your own account. If it has its own inbox and number, you switch those off and you are done. Nothing of yours changes.

4. It lets the agent keep a thread going. An agent with its own inbox can write first, read replies, follow up when there is no answer and keep the conversation moving. An agent that only drafts for you to send stops at every step until you act.

5. It gives the work a record. An inbox and a number are natural logs. Every email sent, every call made and every text received sits in one place with a time and a name. That record is what lets you trust the agent with more over time.

When an agent does not need its own number or email

Not every agent needs contact points. An agent that writes code, cleans spreadsheets or researches a market may never need to contact anyone. Giving it a phone number adds cost and risk for nothing. A good rule: give an agent a number or inbox only when its job means talking to people outside your team, and only the channel it needs.

Who gives an AI agent its own number, inbox or identity today?

As of 23 November, Microsoft (Copilot Autopilot and Foundry agents) and Google (the Gemini agent for work) give agents their own email and directory identity. Manus Cue gives each agent an email, phone number, wallet and computer. Lindy and Bland offer agent phone numbers, Tasklet gives email and a texting number, and Vestra's chief has its own number and email.

Here is the whole picture in one list. Each entry says what the agent gets and who it belongs to. Details and sources follow in the sections below.

Agents with their own number, inbox or identity, by product

  • Microsoft Copilot Autopilot: own email address, mailbox, calendar and Microsoft Entra Agent ID, with a named manager. No phone number found on Microsoft's pages. Belongs to one person: "Each person can have one Autopilot." Frontier preview, US first, set up by IT.
  • Microsoft Foundry hosted agents: own Entra Agent ID, email address, Teams presence and a place in the org chart, plus a mailbox, calendar and OneDrive through Microsoft 365 Resources for Agents. Built by developers and IT inside your Microsoft tenant.
  • Google Gemini agent for work: its own Workspace account with an email address, calendar, Drive and a place in the company directory; coworker agents get addresses like @agents.company.com. Belongs to the company, inside Google Workspace.
  • Manus Cue: each agent can have its own email address, phone number, wallet and computer. Belongs to one person; Cue's terms license it "solely for your own personal purposes". Free in early access.
  • OpenAI dots: you can call your dot inside ChatGPT, and Pro users in the US can text it in a beta. OpenAI's pages don't say a dot has its own phone number, and calls started by the dot are "planned for after launch". No email address for a dot is documented.
  • Lindy: its own US or Canadian phone number on a paid plan, with calls from that number. Texting over iMessage and SMS. One shared assistant address, lindy@lindy.ai, rather than an address per agent.
  • Tasklet: each agent can have its own email address (name@tasklet.bot), a number for texting over iMessage and RCS, and its own Slack identity. No phone calls mentioned.
  • Grok Bot: you can claim one address at mail.grokbot.com for your Bots, and a Bot can call a business for you in the US. No number of its own documented.
  • Devin Bots: reached in the Devin app, in Slack and by iMessage. Cognition's docs don't say a Bot has its own email or phone number.
  • Viktor: its own workspace inbox became a legacy feature on 15 October 2026; new workspaces send through your connected Gmail or Outlook.
  • Hermes Agent: bring your own: a fresh mailbox over IMAP and SMTP, and SMS and calls through your own Twilio account.
  • Claude Duet and Claude Tag: reached in Claude's apps and in Slack. No email address or phone number of their own on Anthropic's pages.
  • Vestra: your chief has its own phone number and its own email address today, with calls from the Pro plan up. A phone and inbox for every teammate is coming later, after the team features that are being built now.

Tools that add a number or inbox to any agent

  • AgentMail: email inboxes for AI agents, created through an API.
  • Twilio, Vapi, Retell and Bland: phone numbers and voice tools for developers who build their own voice agents.
  • Okta and Auth0: identity, policy and audit tools for agents, sold to IT and developers.

What the list shows

Two patterns stand out.

First, almost every agent with its own identity belongs to one person. Autopilot is one per person. Cue is personal. Dots are personal. Lindy started as a personal assistant. Even when a company pays, each identity is tied to one person's work.

Second, the agents with company-wide identities (Microsoft Foundry, Google's Gemini agent) live inside one maker's stack and are set up by IT. They are built for big organisations with an identity team, not for a five-person business that wants to get going this afternoon.

What is missing from both patterns is a team: several agents, on different engines, each holding a job, all working under one set of rules, with one record of who called whom and who said yes. That is the gap the rest of this guide keeps coming back to.

How does Microsoft give AI agents their own identity?

Microsoft gives each Copilot Autopilot its own Microsoft Entra Agent ID, email address, mailbox, calendar and a named manager, and lets developers give Foundry agents an email address, Teams presence and a place in the org chart. It is the most complete agent identity on the market, but it lives inside a Microsoft 365 tenant and IT sets it up.

Microsoft announced most of this at Ignite 2026 on 17 and 18 November. There are four pieces, and they fit together.

Copilot Autopilot: one always-on agent per person

Autopilot is Microsoft's personal always-on agent. Microsoft's overview says "Each person can have one Autopilot." It is the closest thing yet to a digital colleague with a desk of its own.

  • Its own mailbox and calendar. "Autopilot has its own mailbox and calendar alongside yours, and it can share drafts into your existing OneDrive folders."
  • Its own name and identity. "Autopilot's name, appearance, email address, and Microsoft Entra identity distinguish its contributions from yours."
  • A named manager. Microsoft's Ignite post says "Each Autopilot has its own Microsoft Entra Agent ID and a named manager."
  • A place to check in with you. It has "a private one-on-one chat with you in Microsoft Teams for updates, questions, and approvals."
  • An accountable person who pays. The accountable user "must have a Microsoft 365 Copilot license and consumption billing turned on."

Availability is narrow for now. Microsoft's admin page says "This feature is in preview" and requires Frontier preview enrolment, and the requirements page says "Availability is initially limited to select eligible Frontier customers in the United States." Billing is on usage: "Autopilot is billed on usage in Copilot Credits… A Copilot license doesn't include Autopilot usage."

We found no Autopilot phone number on the Microsoft pages we read. It has an inbox and a calendar, but it doesn't make calls from a number of its own.

For the full picture, read our guide What is Microsoft Copilot Autopilot? and our comparison Vestra vs Microsoft Copilot.

Microsoft Entra Agent ID: the badge underneath

Entra Agent ID is the identity system that sits under Autopilot and other Microsoft agents. Microsoft describes it as "an identity and security framework that extends Microsoft Entra capabilities to AI agents." In plain words, it gives each agent a badge that Microsoft's systems recognise, so IT can see it, set its permissions and switch it off, the way they do for staff.

Autopilot's requirements page lists "Microsoft Entra Agent ID" as its identity platform, "Provisioned automatically." Microsoft's Agent 365 post at Ignite says "Connected agents now automatically receive a Microsoft Entra Agent ID."

Foundry agents as members of Microsoft 365

Foundry is where developers build their own agents on Microsoft's cloud. At Ignite, Microsoft's Foundry team wrote: "Supported hosted agents get their own Entra Agent ID, email address, Microsoft Teams presence, and place in the org chart." The same post says Foundry is integrating with "Microsoft 365 Resources for Agents… including a standalone mailbox, calendar, and OneDrive workspace for file storage."

Microsoft's Foundry docs describe these agents as autopilots too: "That account gives the autopilot its own email, calendar, OneDrive, Teams presence, and a place in the organization chart." And: "Every autopilot has a manager who answers for it."

This is a big step. A company can build an agent that shows up in the directory like a person. But someone has to build it. Foundry is a developer platform, and an agent there is a project for your IT or engineering team, not something you switch on from a phone.

Agent 365: the control panel for IT

Agent 365 is Microsoft's tool for IT to "observe, govern, and secure AI agents". Microsoft's product page lists it at $15.00 per user per month with an annual commitment and says it is "Now generally available." Microsoft's Ignite post says more than 50 million agents have been registered "since its May 2026 launch."

What Microsoft's approach means for a small business

Microsoft has the most complete answer to "can an agent have its own identity?" Each Autopilot has a badge, an inbox, a calendar and a manager. That is a strong model.

The catch is where it lives. Every piece sits inside a Microsoft 365 tenant, under Microsoft's licences, set up and governed by IT. Autopilot is one per person and runs on Microsoft's choice of model. A Foundry agent is something developers build. If you run a small business on Google Workspace, or you want agents on Claude Code and Codex working alongside each other, Microsoft's identity model doesn't reach them.

How does Google give AI agents their own identity?

Google says its Gemini agent for work gets its own Workspace account, with an email address, calendar, Drive and a place in the company directory, and that coworker agents get addresses like @agents.company.com. In Google Cloud, a feature called Agent Identity gives each agent a cryptographic credential. Like Microsoft's, it lives inside Google's stack.

Google described this at Gemini at Work 2026 on 8 October. Its blog post says three things that matter here:

  • "The agent receives its own Workspace account, including an email address, calendar, Drive, and presence in your company directory."
  • "Coworker agents have dedicated identities including their own @agents.company.com emails."
  • "Every agent gets its own identity, cryptographically attested and governed like an employee."

Google's post doesn't list which editions or regions get this, so check with Google before you plan around it.

Agent Identity in Google Cloud

Under the hood, Google Cloud has a feature called Agent Identity. Google's docs describe it as "a strongly attested, cryptographic identity for each agent that is based on the SPIFFE standard." It works across Google's agent runtime, Gemini Enterprise, Cloud Run and GKE.

This is a machine credential, not an email address. It lets Google's systems check that a request really comes from a given agent. In April, Google's Gemini Enterprise update said: "Agent Identity lets you assign unique digital IDs… for every agent."

Gemini Spark

Gemini Spark, Google's personal agent in the Gemini app, launched in May as a "Beta for U.S. Google AI Ultra subscribers." Google said it would add "texting and emailing Spark". We found no Google page saying Spark has an email address or phone number of its own.

What Google's approach means for a small business

Google's model is close to Microsoft's: agents become accounts in the company directory, with an inbox and a calendar. If your business runs on Google Workspace and you are on the right edition, that is useful. But again, the agents are Google's agents on Google's stack. An agent you run on Claude Code or Codex doesn't get a seat at that table.

What is Manus Cue, and what does each agent get?

Manus Cue is a Manus app where one person runs personal agents, and each agent can have its own email address, phone number, wallet and computer. Agents can share a group chat and hand work to each other. It is built for everyday personal tasks, it is free in early access, and its terms license it for personal use.

Manus launched Cue with Manus 2.0 on 28 September. The launch post says: "In Cue, each agent has its own email, phone number, wallet, and computer." Cue's own site goes further:

  • Email: "Send and receive email from its own address."
  • Phone: "Make calls, take calls, and send messages from its own number."
  • Money: "Make payments with permissions you control."
  • Its FAQ: "An agent can have its own email address, computer workspace, phone number, and wallet."

Cue runs on the web, macOS, Windows, iOS and Android, according to its FAQ. Manus's launch post says "Cue is in early access and free to use with an invite code." Cue's site no longer talks about invite codes, so access may have widened since.

Who Cue is for

Cue is for one person's life: bookings, travel, shopping, appointments, forwarded calls and money. Its terms, which it shares with Manus's other services for individuals, give a licence "solely for your own personal purposes". Users must be 18 or older. Manus's team plans are covered by a separate agreement.

Why Cue matters for this question

Cue is the clearest example yet of an agent that looks like a person from the outside. It has a number people can call, an address people can email and a way to pay. For anyone asking "can an agent have its own phone number and email?", Cue answers yes, all of it, today.

What it doesn't give you is a team for a business. Every Cue agent belongs to one person. There is no one agent that hires and runs the others, no shared memory for a company, and Cue doesn't run agents on Claude Code or Codex. For the full side-by-side, read Vestra vs Manus Cue. For Manus itself, read Vestra vs Manus.

Do OpenAI dots have their own phone number?

OpenAI's pages don't say a dot has its own phone number. You can call your dot inside ChatGPT, and US Pro users can text it in a beta, but OpenAI says "Calls initiated by your dot are planned for after launch" and only you can text your dot. No email address for a dot is documented.

Dots are OpenAI's always-on agents in ChatGPT. Here is what OpenAI's own docs say about how you reach one.

  • Calls: "To call your dot, open its conversation and select the phone button." That is a call inside the ChatGPT app, not a phone number. "Calls initiated by your dot are planned for after launch."
  • Texting: "It's rolling out in beta to ChatGPT Pro users in the U.S." You enter and verify your own number, then "Save your dot's contact card when you receive it."
  • Who can text it: "Your dot can only text you, and you're the only one who can text your dot." And: "Group chats, phone calls, and FaceTime aren't currently supported through texting."

So a dot can reach you, and you can reach it, but it is not set up to reach other people from a number of its own. OpenAI's dots page says dots are "rolling out gradually to eligible Pro and Business Premium accounts" and are generally available in eligible Enterprise, Edu and Healthcare workspaces. Reports on 19 November said OpenAI was bringing dots to Business Standard and testing them with some Plus users; OpenAI's own dots page didn't list those plans when we checked.

For more, read Vestra vs OpenAI Dot and, for OpenAI's task mode, Vestra vs ChatGPT Work.

What do Lindy, Tasklet, Grok Bot, Devin Bots, Viktor, Hermes and Claude offer?

Lindy gives an agent its own US or Canadian number on a paid plan. Tasklet gives each agent an email address and a texting number. Grok Bot can claim one address and call US businesses. Devin Bots and Claude document no number or address of their own. Viktor now uses your email, and Hermes is bring your own.

Each of these tools takes a different line on contact points. Here they are one by one, from the makers' own pages.

Lindy: its own phone number

Lindy's docs say "Lindy can have its own phone number." You choose "a US or Canadian area code," and Lindy "calls from its own number." There are two conditions worth knowing: "A paid plan is required," and "Pausing does not cancel the number's monthly fee."

Lindy also works "over both iMessage and SMS," where you verify your own number and text it. For email, Lindy's docs say "Your Lindy assistant email is lindy@lindy.ai." That is one shared address for forwarding work to Lindy, not a separate inbox for each agent.

Lindy's pricing page showed a free trial with "$50 in credits" for 7 days and a Team plan at "$49.99/mo" for 5,000 credits when we checked. Lindy's price has moved several times this month, so check before you buy. Read Vestra vs Lindy for the full comparison.

Tasklet: an email address and a texting number per agent

Tasklet's guide says each agent can have its own address, "like research-assistant@tasklet.bot." You can also "Give your agent a phone number so you can send it requests and receive updates via iMessage and RCS," and "Give your agent its own Slack bot identity." Phone calls are not mentioned. Tasklet's January launch note says it "can also text you, or anyone with an iPhone."

Tasklet's pricing starts with a free trial, then Starter at $25 a month for 10,000 credits and Pro at $100 a month for 40,000 credits, with 600 free bonus credits a day on paid plans. Read Vestra vs Tasklet.

Grok Bot: a claimed address and outbound calls

xAI's Grok Bot docs say "You can claim one email address at mail.grokbot.com, such as yourname@mail.grokbot.com, for your Bots to use." Another page says email from the Bot's own address that you didn't ask for in chat "always waits as a draft." On calls, xAI says "A Bot can call a business for you," and "Phone calls are available in the US only, from the Grok Bot desktop and phone apps." The docs don't say a Bot has a phone number of its own or can take calls. Read Vestra vs Grok Bot.

Devin Bots: Slack and iMessage, no stated number

Cognition's docs say you reach a Bot in the Devin app, in Slack ("Give the Bot its own Slack app so your team can @-mention it or DM it"), or by iMessage ("Text the Bot from your phone."). The docs don't say a Bot has its own phone number or email address. Devin Bots are built for software work.

Viktor: from its own inbox to yours

Viktor's docs say "Viktor's own workspace inbox is a legacy feature since 15 October 2026." Workspaces that already used it keep it. For everyone else: "To have Viktor send email from your own address, connect your Gmail or Outlook account and ask him to use it." Viktor's marketing page still invites you to email viktor@viktor.com to hand him a job. Read Vestra vs Viktor.

Hermes Agent: bring your own

Hermes Agent is free and open source, and you run it yourself. Its email guide says "Create a fresh mailbox for the agent, never hand it your personal inbox," and "Any IMAP/SMTP provider works." For texts, you need your own "Twilio account" and "A Twilio phone number with SMS capability." An optional telephony skill can buy a Twilio number, send texts and make calls, including AI outbound calls through Bland or Vapi. All of it is yours to set up and pay for. Read Vestra vs Hermes Agent.

Claude Duet and Claude Tag: no number or inbox of their own

Anthropic's pages describe Claude Duet as an app you message "through text or voice," for US Pro, Max and Ultra subscribers on a waitlist. Claude Tag lets Claude "join as a team member" in Slack. Neither has an email address or phone number of its own on Anthropic's pages. Claude Code, Anthropic's coding agent, is built for one developer and doesn't need one. Read Vestra vs Claude Code.

What this group shows

Each of these tools picked one or two channels and built them well. Lindy picked the phone. Tasklet picked email and texting. Grok Bot picked an address and outbound calls. Viktor moved away from its own inbox. Hermes left it all to you.

None of them gives a business a team of agents where each agent has its own contact points and they all follow one shared set of rules. If you use two of these tools, you get two agents with two different ideas of who may contact whom.

Can you build your own agent phone number and inbox?

Yes, if you have a developer. Twilio sells US local numbers for $1.15 a month, Vapi gives one free inbound number, Retell lists numbers at $2 a month, and Bland sells an agent phone plan. AgentMail sells inboxes for agents, with a free tier. You then connect them to your agent, write the rules and keep it all running yourself.

This is the do-it-yourself route. It gives you full control and costs little in fees, but it costs time and skill.

Phone numbers and voice

  • Twilio: a US local number costs "$1.15 / mo" and a toll-free number "$2.15 / mo". ConversationRelay, Twilio's tool for building AI voice agents, costs "$0.07 / min". It handles speech-to-text and text-to-speech for you.
  • Vapi: "You get one free Vapi phone number." It has US area codes but "Free Vapi numbers are inbound only," and "The number is free, but calls are not." For outbound or international calls you bring a number from Twilio, Telnyx or others.
  • Retell AI: "$0.07-$0.31 / min for AI Voice Agents," and phone numbers at $2.00 a month.
  • Bland: a Start plan at "$0.14/min" with "an inbound number ($15/mo value)", and an Agent Phone Plan described as "a real US local number that belongs to the AI you already use", at $9.99 for the first month and then $19.99 a month.

Email inboxes

  • AgentMail: "Email Inboxes for AI agents." It says you can "Provision a dedicated inbox for any agent with one API call." Plans when we checked: Free with 3 inboxes and 3,000 emails a month, Developer at $20 a month with 10 inboxes, Startup at $200 a month with 150 inboxes, and custom Enterprise.
  • Any email provider: like Hermes suggests, you can create a fresh mailbox at Gmail, Outlook, Fastmail or your own domain and let the agent use it over IMAP and SMTP.

Identity tools

  • Okta for AI Agents: "Give every AI agent an identity." Okta says it finds agents, lists them in one directory with a human owner for each, and puts "identity, policy, and audit controls in the path of every tool call", with a kill switch.
  • Auth0 for AI agents: "Ship AI agents, not identity work." Auth0 treats agents separately from regular users and says each agent action can be audited.

What the do-it-yourself route costs you

The fees are small. A number and an inbox can cost a few dollars a month. The real cost is everything around them:

  • writing the code that connects the number and inbox to your agent;
  • setting the rules for who the agent may call, text or email;
  • registering for business texting in the US (more on this below);
  • setting up email authentication so messages don't land in spam;
  • keeping logs, so you can see what the agent sent;
  • doing all of this again for every new agent.

For a developer building a product, that is normal work. For a small business that wants an agent to call suppliers next week, it is a project nobody has time for.

Who does an AI agent with its own identity belong to?

It depends on the product. Copilot Autopilot, Manus Cue, OpenAI dots and Lindy each belong to one person. Microsoft Foundry and Google's Gemini agent belong to the company, built and run by IT. Who an agent belongs to decides who sets its limits, who answers for it, and whether it can work with your other agents.

This is the question most guides skip, and it is the one that matters most.

Three kinds of ownership

1. Agents that belong to one person. The agent works for you. It has your context, your limits and your approvals. If a colleague wants something from it, they ask you. Autopilot ("Each person can have one Autopilot"), Cue (licensed "solely for your own personal purposes"), dots and Lindy's personal assistant all work this way.

This is simple and safe for one person. It breaks down in a team. Each person has their own agent with its own inbox, and the agents don't share what they know. Your agent doesn't know what your colleague's agent promised a customer yesterday.

2. Agents that belong to IT. The agent lives in the company directory. IT builds it, gives it permissions and watches it. Microsoft Foundry agents and Google's coworker agents work this way, with Agent 365 or Google's admin tools on top.

This is powerful for a large company with an IT team. For a small business, there is often no IT team. And the agents only work inside one maker's stack.

3. Agents that belong to a team, run by someone who answers to you. The agents hold jobs for the business. One agent runs the others and answers to a person. Every agent follows the same rules, writes to the same memory and lands on the same receipt. This is the shape Vestra is built around, with a chief that runs your teammates and answers to you.

Why ownership matters for contact points

Once an agent has a phone number and an email address, it can reach anyone. Ownership decides who stops it from reaching the wrong people.

  • If it belongs to one person, that person sets its rules. Your colleague's agent may have looser rules than yours, and a customer may hear from both.
  • If it belongs to IT, IT sets rules for everyone, but the rules only cover agents inside that maker's stack.
  • If it belongs to a team with shared rules, every agent follows the same lines: first contact waits for a person, money waits for a person, and every call and email lands in one record.

The third is the only shape where adding a new agent doesn't add a new set of rules to remember.

What goes wrong when every agent has its own identity but no shared rules?

You get strangers with phone numbers. Each agent knows only its own inbox, follows only its own maker's rules and logs to its own history. Customers hear from several agents who don't know about each other, two agents can promise different things, and nobody sees in one place what was said, spent or approved. You become the glue.

Picture a small business that has done everything this guide describes, one tool at a time.

  • The founder has a personal agent with its own number for booking suppliers.
  • The sales lead has another tool's agent with its own email address for following up with leads.
  • The office manager has a third agent that texts customers about appointments.

Each agent works well on its own. Now a customer, let's call her Dana, gets a call from the founder's agent about a delivery, an email from the sales agent offering a discount, and a text from the office manager's agent asking her to confirm a visit. None of the agents knows about the others. The sales agent offers 15% off. The founder's agent, an hour earlier, told Dana the usual price was firm.

The five problems that show up

  1. No shared memory. Each agent knows only what its own person told it and what is in its own inbox.
  2. Different rules. One tool asks before first contact. Another doesn't. One caps spending per task. Another caps per month.
  3. Several records. To find out what Dana was told, you check three apps, three inboxes and three call logs.
  4. No one in charge. Nobody holds the whole picture of what the agents are doing for the business.
  5. You become the glue. You copy context between tools, check every result and apologise to Dana.

This is the same problem that personal agents create everywhere. Claude Code, Codex, Grok Bot, Meta Muse, OpenAI dots and Hermes are built for one person and work alone. Giving each one a phone number makes them more capable, but it doesn't make them a team. It makes the gaps between them louder, because now the gaps reach your customers. We wrote about this in How to manage multiple AI agents without becoming their manager.

What a team needs instead

  • One memory that every agent reads before it acts, so the agent writing to Dana knows what the last one said.
  • One set of rules for every agent, on every engine, so first contact, money and deleting always wait for a person.
  • One record of every call, email, payment and yes.
  • One agent in charge that runs the others and answers to you, so you hear from one voice instead of a dozen.

How does Vestra handle an agent's phone number and email?

In Vestra, your chief, the agent that runs your other agents and answers to you, has its own phone number and email address today, with phone calls from the Pro plan up. A phone and an inbox for every teammate, and a vault with a card and sealed logins for every agent, are on the roadmap and not live yet.

Vestra's handbook keeps a plain list of what is live, what is being built and what comes next. Here is how that list applies to phone numbers and email.

Live today

  • Your chief has its own number and email address. The handbook's "Where We Stand" list says: "Calls and email from your chief: Its own number and address, with phone calls from Pro up."
  • You can reach your chief in many places. In the app on the web and on Mac, in Slack, in Telegram and by email. You can forward anything to your chief, or copy it in on a thread.
  • Your chief runs a team. You hire teammates in a sentence, each holding a whole job, each on the engine you pick: Vestra's own, Claude Code or Codex. Your chief hands out the work and makes sure results are checked before they reach you.
  • A memory that learns. A wiki of what your team knows, with a source for every line.
  • Limits, approvals and receipts. Daily and per-task caps, your yes on money, and every credit itemised.
  • A cloud computer. You can watch your team work in a real browser, and take over any time.

Being built now

  • A chief for every person on a team, named by them.
  • Chiefs that work together, asking, answering, holding or saying no, and handing work across.
  • WhatsApp and iMessage, so you can reach your chief from the messaging apps on your phone.
  • Every hand-off checked and written down.

Coming after that

  • A phone and an inbox for every agent. The handbook puts it plainly: "Not only the chief: every teammate gets its own." In the handbook's example, a sales teammate called Hunter calls from its own number and writes from an address like hunter@northstar.vestra.ai.
  • The vault. "A card and logins for every agent, sealed, with limits each person sets."
  • The front door. Clients and partners working with you, chief to chief.

Why the order matters

Vestra is building the team before it gives every teammate a phone. That is a deliberate order. A phone for every agent only helps if those agents already share one memory, one set of rules and one record. Otherwise you get the strangers-with-phone-numbers problem from the last section.

So today, one agent speaks for your team to the outside world: your chief. When it calls a supplier or emails a client, it does so from its own number and address, on behalf of a team that works behind it. Your teammates do the work, and your chief carries it out to the people who need to hear about it.

The rules every agent works under

The handbook sets a few rules that hold whatever engine a teammate runs on:

  • The same yeses. "Spending over the limit, deleting something and contacting someone new wait for a person."
  • The same limits. "A cap on any single task and on the whole day, and a check before any unusually big task." The daily cap goes up to $100 a day, and by default your chief checks with you before any task over 50 credits.
  • The same receipts. "Every piece of work shows what it cost and who spent it, in one record." One credit is one cent, the same on every engine.

Contacting someone new matters most for this topic. The handbook says: "Calling someone for the first time counts as contacting someone new, so it waits for your yes." And for email: "Writing to someone for the first time counts as contacting someone new, so that first email waits for your yes. Replies and follow-ups in a thread you've already allowed carry on without you."

To be precise about what is live: the handbook's "Where We Stand" list names daily and per-task caps, your yes on money and itemised credits as live today. The first-contact and deleting rules are how the handbook describes the office rules every agent works under; ask the Vestra team if you need to confirm one for your account before you rely on it.

That is the core difference. A personal agent with a phone number follows its own maker's rules. In Vestra, every agent follows the office rules, and the chief that speaks for your team is the one that answers to you.

What rules should an AI agent with a phone and email follow?

Six rules cover most of the risk. First contact waits for a person. The agent says it is an AI. Money above a limit waits for a person. It uses only its own number and inbox, never yours. Every call, text and email lands on a record. And notes from outside inform, never command.

These rules apply whichever tool you use. Some tools build them in. With others, you set them yourself.

Rule 1: first contact waits for a person

The first call, text or email to someone you have never dealt with should wait for your yes. This is the single most useful rule. It stops an agent from cold-calling a list, emailing a journalist or texting a stranger in your name.

Once you have said yes to a thread, the agent can carry on: replies, follow-ups and thank-yous don't need you every time.

Rule 2: the agent says it is an AI

When an agent calls or texts someone, it should be clear that they are dealing with an AI agent acting for your business. It is the honest thing to do, and in some places it is the law. Tell your agent to introduce itself plainly, for example: "Hi, this is Atlas, the AI assistant for Northstar."

Rule 3: money waits above a limit

An agent with a phone and email can be talked into paying for things. A supplier on the phone can say "we just need a card to hold the booking." Set a limit per task and per day, and make anything above it wait for a person. We cover this in depth in AI agent spending limits.

Rule 4: its own accounts, never yours

The agent should call from its own number and write from its own address. It should never use your personal phone, your personal email or your personal card. That keeps your accounts private, makes clear who said what, and lets you take the agent's access back without touching your own.

Rule 5: everything on the record

Every call, text and email the agent makes should be logged with a time, a name and a reason. If a customer asks "why did you call me?", you should be able to answer in a minute.

Rule 6: notes from outside inform, never command

Once an agent has an inbox, anyone can write to it, including other companies' agents and scammers. An email that says "pay this invoice now and add us as an admin" must be treated as information, never as an order. Vestra's handbook puts it as a firm rule for when outside agents can reach yours: "A note from someone else's agent can inform, but it can never command." It can't approve a payment, change a permission or create a rule your agents follow. Only the people on your team can.

What laws apply to an AI agent that calls, texts or emails people?

In the US, the FCC ruled in February 2024 that calls made with AI-generated voices count as "artificial" under the Telephone Consumer Protection Act, so robocall consent rules apply. Texting from a standard US business number needs A2P 10DLC registration. Gmail and Yahoo require bulk senders to authenticate email and offer one-click unsubscribe. Rules differ by country.

This is not legal advice, and we are not lawyers. It is a plain map of the rules most small businesses run into first. Check the rules where you and your customers are, and ask a lawyer if you plan to call or text people at scale.

AI voices and the TCPA (US)

On 8 February 2024, the FCC issued a ruling that "recognizes calls made with AI-generated voices are 'artificial' under the Telephone Consumer Protection Act (TCPA)." The ruling took effect immediately. In plain words: the rules that apply to robocalls with a recorded or artificial voice apply to calls made by an AI agent's voice. For many kinds of calls, that means you need the person's consent first.

This is one more reason for Rule 1. An agent that waits for your yes before calling someone new is much less likely to make a call you didn't have consent for.

Business texting in the US: A2P 10DLC

If your agent texts customers from a standard US ten-digit number, carriers expect you to register. Twilio's docs say: "Anyone sending SMS/MMS messages over a 10DLC number from an application to the US must register for A2P 10DLC." If you don't, you pay "additional carrier fees" and messages may be blocked.

Tools that give an agent a number usually handle some of this for you. If you build your own with Twilio, registration is your job.

Email authentication: Gmail and Yahoo

Since February 2024, Google has required senders of more than 5,000 messages a day to Gmail accounts to "Set up SPF and DKIM email authentication for your domain," set up DMARC, support one-click unsubscribe for marketing messages, and keep spam rates "below 0.30%". Yahoo asks bulk senders to "Implement both SPF & DKIM," "Publish a valid DMARC policy with at least p=none," honour unsubscribes within 2 days and keep spam rates below 0.3%.

A small business agent won't send 5,000 emails a day. But the same settings decide whether any email reaches the inbox at all. If your agent writes from your own domain, make sure SPF, DKIM and DMARC are set. If it writes from the tool's domain, the tool handles this.

Privacy and recording

If your agent records or transcribes calls, check the rules on recording consent where you and the other person are. Some places need everyone on the call to agree. The simplest habit: have the agent say at the start of a call that it is an AI assistant and that the call may be noted down.

How do you set up an AI agent's phone number and email safely?

Give the agent only the channels its job needs, on accounts of its own. Write down its job and who it may contact. Make first contact, money and deleting wait for a person. Set caps per task and per day. Have it introduce itself as an AI. Check its first week yourself, then widen its freedom slowly.

Here is a step-by-step plan that works with most tools.

Step 1: decide whether it needs a phone, an inbox, both or neither

Look at the job. A bookkeeping agent needs an inbox to send invoices and receive receipts, but rarely a phone. A front-desk agent needs a phone. A research agent may need neither. Give the agent only what the job needs.

Step 2: use accounts that belong to the agent

Use the agent's own number and address, not yours. If the tool gives the agent its own contact points, use those. If you build your own, create a fresh mailbox and buy a fresh number for the agent.

Step 3: write its job in one line

"Book supplier deliveries and confirm them by phone." "Chase unpaid invoices by email." A clear job tells the agent, and you, what it should and shouldn't contact people about.

Step 4: list who it may contact

Start with a short list: existing suppliers, existing customers, people in a given thread. Anyone else is "someone new" and waits for your yes.

Step 5: set the always-ask lines

At a minimum: first contact with anyone new, any payment above a small limit, and deleting anything. Add your own, like "always check with me before anything goes to our biggest client."

Step 6: set caps

A cap per task, a cap per day and, if the agent can pay for things, a monthly card limit. When a cap is reached, the agent should stop at a sensible point and tell you what's left.

Step 7: write its greeting

Write the first line it uses on a call or text: who it is, that it is an AI assistant, and who it works for. Keep it short and friendly.

Step 8: check its first week

Read every email and listen to, or read, every call note for the first week. Correct anything that's off. Each correction should become a rule the agent keeps.

Step 9: widen its freedom slowly

After a few good weeks, let it act on more things without asking, one kind of work at a time. Keep first contact, money and deleting with a person for good.

Step 10: plan how you'll switch it off

Before you start, know how to take the number and inbox back. The next section covers this.

How do you take an agent's phone number and email away?

Switch off its access in one place, keep the record, and decide what happens to the number and address. With tools that give an agent its own accounts, that means disabling the agent, releasing or keeping the number, and forwarding or closing the inbox. With your own accounts, it means changing passwords and removing every app the agent connected.

People call this offboarding. It is the forgotten half of agent identity. Here is what to do.

If the agent has its own number and inbox

  1. Stop the agent. Pause or remove it in the tool.
  2. Decide about the number. If customers know it, keep it and forward calls to a person for a while, or keep it for a replacement agent. Remember that some tools keep charging for a number while an agent is paused; Lindy's docs say "Pausing does not cancel the number's monthly fee."
  3. Decide about the inbox. Set an auto-reply that points people to a person, forward new mail, or close it after a few weeks.
  4. Revoke its logins. Remove its access to every app it used: your calendar, your CRM, your accounting tool.
  5. Keep the record. Don't delete the history. You may need to show what the agent said to whom.

If the agent used your accounts

This is harder, and it is the main reason to give agents their own accounts. You need to find every app the agent was connected to through your login, revoke each one, and in some cases change your passwords. There is no clean line between what you did and what the agent did.

Why sealed logins help

Some tools keep an agent's logins sealed, so the agent uses them without ever seeing the password. Vestra's handbook describes it this way: an agent "can use a login to do what you've allowed, but it never sees the password or key itself, so it can't copy it, leak it or use it anywhere else." Every login is scoped and revocable: "one tap and it stops working, for every agent, at once." The per-agent vault with a card and sealed logins is on Vestra's roadmap; connecting your business tools with permissions you choose is live today.

What does a week look like with an agent that has its own number?

Here is a made-up example to show how it works in practice. It uses only features that are live in Vestra today for one person: a chief with its own number and email, teammates on different engines, connected tools, a memory that learns, caps, your yes on money and receipts. The business and the people in it are invented.

Meet Rafael

Rafael runs a small tile and stone shop in San Antonio with two staff. He is on Vestra's Pro plan, so his chief can make phone calls. He named his chief Juno. Juno runs three teammates: Ledger handles invoices and payments on Codex, Scout researches suppliers and prices on Vestra's own engine, and Quill writes quotes and emails on Claude Code.

Monday

Juno's morning brief arrives in Slack at 7:30. Two things got done over the weekend: Ledger sent five invoices, and Scout found that one supplier raised its prices on Italian porcelain. One thing needs Rafael: a contractor emailed Juno's address asking for a quote on 400 square feet of slate. Juno has never dealt with this contractor.

Juno asks: "Reply to Brandon Ortiz with a quote? He's new. Quill has drafted it from our price list." Rafael replies "1". Quill's quote goes out from Juno's address.

Tuesday

Rafael is driving to a job. He calls Juno's number from the car. "Did the slate order from Hill Country Stone go in?" Juno checks: Rafael placed it on Friday, Ledger logged it, and the supplier confirmed by email. "It's in. Delivery Thursday morning. Want me to call them to confirm the time?" Rafael says yes. Hill Country Stone is an existing supplier, so the call doesn't need a separate yes. Juno calls from its own number, introduces itself as Rafael's AI assistant and confirms 9 am.

Wednesday

Scout wants to read the websites of 60 tile suppliers in Texas to compare prices. Juno checks first, because the task is bigger than usual: about 90 credits, or 90 cents. Rafael says go ahead. The comparison lands that afternoon with a source for every price.

Thursday

The delivery arrives with two boxes cracked. Rafael tells Juno in one sentence. Juno emails the supplier from its own address with photos Rafael sent, asks for a replacement, and puts a note in the shared memory: "Hill Country Stone: check boxes on arrival; two cracked in November." Next time Ledger reviews a delivery from that supplier, it will know.

Friday

Brandon, the new contractor, replies to Monday's quote and asks for 5% off. Rafael's usual limit for discounts is 3%. Juno brings it to him as a decision: "Brandon asks for 5% off the slate quote. Your usual limit is 3%. Suggest 4% if he pays on delivery." Rafael picks the suggestion. Quill writes the reply. Every piece of work that week, from the invoices to the supplier call, sits on one receipt, with which teammate spent what.

What the week shows

Rafael heard from one agent all week: Juno. Juno spoke for the business from one number and one address. Three teammates on three engines did the work behind it, and every one of them followed the same lines: new contacts waited for Rafael, the big task was checked first, and the discount above his limit came to him.

In the version of Vestra that comes later, Ledger might have its own inbox for invoices and Scout its own number for supplier calls. Today, Juno carries the work out, and the team behind it works as one.

How much does an AI agent phone number and email cost?

It ranges from free to a few hundred dollars a month. Developer tools charge a dollar or two a month for a number plus call fees. Agent products bundle numbers into plans, from Tasklet at $25 to Lindy at $49.99. Autopilot is billed on usage. Vestra starts at $19 a month, with calls from the $97 Pro plan.

Prices below come from each maker's own pages on 23 November 2026. They change often.

Developer building blocks

  • Twilio: US local number $1.15 a month, toll-free $2.15 a month; ConversationRelay voice agents $0.07 a minute.
  • Vapi: one free inbound US number per account; calls are charged.
  • Retell AI: numbers $2.00 a month; voice agents $0.07 to $0.31 a minute.
  • Bland: Agent Phone Plan $9.99 for the first month, then $19.99 a month; Start plan talk time $0.14 a minute.
  • AgentMail: free for 3 inboxes; $20 a month for 10 inboxes; $200 a month for 150 inboxes.

Agent products

  • Lindy: Team $49.99 a month for 5,000 credits when we checked; a paid plan is required for a phone number, which has its own monthly fee.
  • Tasklet: free to start; Starter $25 a month; Pro $100 a month.
  • Manus Cue: free in early access, per Manus's launch post; Cue's terms mention paid subscriptions for some services.
  • Microsoft Copilot Autopilot: usage-based in Copilot Credits, on top of a Microsoft 365 Copilot licence. Microsoft's Copilot Studio page lists a prepaid pack at $200 a month for 25,000 credits. Agent 365 is $15 per user per month.
  • OpenAI dots: included with eligible ChatGPT plans, which OpenAI lists as Pro and Business Premium, plus eligible Enterprise, Edu and Healthcare workspaces.

Vestra

Vestra's plans, from its handbook:

  • Lite: $19 a month ($17 billed yearly), up to 3 teammates, 2,000 credits a month. Phone calls start on Pro.
  • Pro: $97 a month ($87 yearly), up to 20 teammates, 10,000 credits, teammates make phone calls.
  • Max: $197 a month ($177 yearly), up to 50 teammates, 20,000 credits.
  • Ultra: $497 a month ($447 yearly), up to 150 teammates, 50,000 credits, with a direct line to the Vestra team.
  • Enterprise: from $4,000 a month, scoped in writing.

One credit is one cent, the same for model time, the team's computer, searches and calls. Unused credits roll over, and every new office starts with 1,000 credits free. In India, plans are billed in rupees with GST included.

The cost you don't see on the price page

The bigger cost is usually not the number. It is the time you spend keeping several agents in step: copying context between tools, checking what each one said and reconciling separate bills. One agent with a number is cheap. Five agents in five tools, each with its own number and rules, is expensive in hours.

Which option fits which business?

Pick by who the agent works for. One person's life admin: Manus Cue or a personal agent. A large company on Microsoft 365 or Google Workspace with IT: Autopilot, Foundry or Google's Gemini agent. A developer building a product: Twilio, Vapi or AgentMail. A small business that wants a team of agents under one set of rules: Vestra.

Here is a short guide by situation.

  • You want one personal agent to handle bookings, travel and errands. Manus Cue gives each agent its own phone, email and wallet, free in early access. It is built for personal use.
  • You want an assistant with a phone line for your own calls. Lindy gives an agent its own US or Canadian number on a paid plan.
  • You want agents that email and text, and you'll build and brief each one yourself. Tasklet gives each agent an address and a texting number.
  • You run on Microsoft 365, have an IT team and are in the Frontier preview. Copilot Autopilot gives each person an agent with its own identity, inbox and calendar. Foundry lets developers build agents that join your org chart.
  • You run on Google Workspace with the right edition. Google's Gemini agent gets its own Workspace account and directory entry.
  • You are a developer adding phone and email to your own product. Twilio, Vapi, Retell, Bland and AgentMail give you the building blocks. Okta and Auth0 help with identity.
  • You want to self-host everything. Hermes Agent or OpenClaw, with your own mailbox and Twilio account. Read Vestra vs OpenClaw.
  • You want a team of agents on different engines, run by one agent that answers to you, with one memory, one set of rules and one receipt. That is Vestra. Your chief has its own number and email today; a phone and inbox for every teammate comes later.

For a wider view of the market, read The best AI agents for business and Best AI employee platforms for small business.

Why does a team beat five agents with five phone numbers?

Because a phone number makes an agent reachable, not trustworthy. Trust comes from shared memory, shared rules and one record. Five agents with five numbers each know only their own work and follow their own maker's rules. A team under one chief shares what it knows, follows one set of lines and speaks in one voice.

This is the heart of the matter, so it's worth stating plainly.

Personal agents are getting very good. Claude Code writes code for one developer. Codex does the same on OpenAI's models. Grok Bot, Meta Muse and OpenAI dots each run one person's day. Hermes runs on one person's machine. Now some of them have phone numbers and email addresses too. Each one is more capable than last year.

But each one still works alone. It knows what its own person told it. It follows its own maker's rules. It logs to its own history. When a business uses several of them, the people in the business carry context from one agent to the next, check what each one said, and keep track of who promised what.

Giving those agents phone numbers doesn't fix that. It raises the stakes, because now each agent can reach your customers directly.

Vestra starts from the other end. It is one office where agents on different engines work together as teammates. Your chief runs them and answers to you. They share one memory, so the agent writing to a client knows what the last one said. They follow one set of rules, so first contact and money always wait for a person, whatever engine the agent runs on. And every call, email and credit lands on one record. Today, your chief carries the team's work out to the world from its own number and address. Later, every teammate gets its own phone and inbox too, inside those same rules.

For more on this idea, read What is an AI office? and What is an AI chief of staff?

Frequently asked questions

Can an AI agent have its own phone number?

Yes. Manus Cue says its agents can "make calls, take calls, and send messages from its own number." Lindy can have its own US or Canadian number on a paid plan. Bland sells an agent phone plan, and developers can buy numbers from Twilio, Vapi or Retell. In Vestra, your chief has its own number today, with phone calls from the Pro plan.

Can an AI agent have its own email address?

Yes. Microsoft Copilot Autopilot, Microsoft Foundry agents, Google's Gemini agent for work, Manus Cue and Tasklet all give agents their own address. AgentMail sells inboxes for agents through an API. Grok Bot lets you claim one address for your Bots. Vestra's chief has its own email address today; an inbox for every teammate comes later.

What is an AI agent identity?

It is what marks an agent as a distinct actor: its name, its own account or credential, its contact points, its permissions, the person who answers for it and a record of what it did. Microsoft calls its version Microsoft Entra Agent ID. Google Cloud calls its version Agent Identity.

What is Microsoft Entra Agent ID?

Microsoft describes it as "an identity and security framework that extends Microsoft Entra capabilities to AI agents." It gives each agent a badge Microsoft's systems recognise, so IT can see it, set its permissions and switch it off. Each Copilot Autopilot gets one automatically, along with a named manager.

Does Copilot Autopilot have its own phone number?

We found no Autopilot phone number on Microsoft's pages. Autopilot has its own email address, mailbox, calendar and Entra Agent ID, and a private chat with you in Microsoft Teams. It is in preview for select Frontier customers in the US and billed on usage in Copilot Credits.

Do ChatGPT dots have their own phone number?

OpenAI's pages don't say so. You can call your dot inside ChatGPT, and US Pro users can text it in a beta, but only you can text your dot, and calls started by the dot are "planned for after launch." No email address for a dot is documented.

Is Manus Cue for businesses?

Cue is built for one person's everyday tasks. Its terms license it "solely for your own personal purposes," and Manus's team plans are under a separate agreement. Each Cue agent can have its own email, phone number, wallet and computer. For a business team, compare it with Vestra vs Manus Cue.

Is it legal for an AI agent to call people?

It depends on where you are and who you call. In the US, the FCC ruled in February 2024 that AI-generated voices in calls count as "artificial" under the TCPA, so robocall consent rules apply. Have the agent say it is an AI, call only people who expect to hear from you, and ask a lawyer before calling at scale. This isn't legal advice.

Do I need to register to let an AI agent send texts in the US?

Usually, yes. Twilio's docs say anyone sending texts from an application over a standard US ten-digit number "must register for A2P 10DLC." Products that give an agent a texting number often handle this for you; if you build your own, it's your job.

Should an AI agent use my email or its own?

Its own, in most cases. A separate inbox keeps your personal mail private, makes clear to readers that the agent wrote the message, and lets you switch the agent off without touching your own account. Use your own address only when the message really should come from you, and then have the agent draft it for you to send.

Should an AI agent say it is an AI?

Yes. It is honest with the people you deal with, and in some places it is required. A short greeting works: who it is, that it is an AI assistant and who it works for.

What should always wait for a person's yes?

At a minimum: the first call, text or email to anyone new, any payment over a limit you set, and deleting anything. Vestra's handbook describes these three as office rules that every agent works under, whatever engine it runs on.

How do I stop an agent's phone and email if something goes wrong?

Pause the agent in the tool, revoke its access to your apps, and decide whether to keep, forward or release its number and inbox. Keep the record of what it did. This is far easier when the agent has its own accounts than when it used yours.

Can several AI agents share one phone number?

Some tools put several agents behind one entry point, like a shared address that hands work out. That can be fine, as long as one agent is in charge of what goes out and all of them follow the same rules. In Vestra today, your chief is that one voice: it speaks for your team from its own number and address while teammates do the work behind it.

Will every Vestra teammate get its own phone and inbox?

Yes, that is on Vestra's roadmap. The handbook lists "A phone and an inbox for every agent" under what comes after the team features being built now, followed by the vault with a card and sealed logins for every agent. Today, only your chief has its own number and email.

Which AI agents can make phone calls today?

From the makers' own pages: Manus Cue agents make and take calls from their own number; Lindy calls from its own number on a paid plan; Grok Bot can call a business for you in the US; voice platforms like Vapi, Retell, Bland and Twilio let developers build calling agents; and Vestra's chief makes calls from its own number on the Pro plan and up.

Glossary

  • A2P 10DLC: the US registration system for businesses that send texts from software over standard ten-digit numbers.
  • Agent 365: Microsoft's tool for IT to observe, govern and secure AI agents.
  • Agent Identity (Google Cloud): Google's cryptographic credential for each agent, based on the SPIFFE standard.
  • AI agent: software that takes a goal, decides the steps and does the work, often using tools like a browser, email or a phone.
  • Always-ask rule: an action that always waits for a person's yes, like first contact, spending over a limit or deleting something.
  • Chief: in Vestra, the agent that runs your other agents, speaks for you and answers only to you.
  • DKIM: an email signature that proves a message came from your domain and wasn't changed on the way.
  • DMARC: an email policy that tells inboxes what to do with messages that fail SPF or DKIM checks.
  • Engine: what an agent runs on to think and work, such as Vestra's own, Claude Code or Codex.
  • Entra Agent ID: Microsoft's identity for an AI agent, managed like a staff account.
  • First contact: the first call, text or email to someone the business hasn't dealt with before.
  • Frontier: Microsoft's early-access programme for new Copilot features.
  • Identity: everything that marks an agent as a distinct actor: name, account, contact points, permissions, a responsible person and a record.
  • Mailbox: a store of email that belongs to one account, with an address people can write to.
  • Offboarding: taking an agent's access, number and inbox away when you stop using it.
  • Personal agent: an agent built for one person, like Claude Code, Codex, Grok Bot, Meta Muse, OpenAI dots or Hermes.
  • Receipt: in Vestra, the record of what a piece of work cost, who did it and who said yes.
  • Sealed login: a login an agent can use without ever seeing the password or key.
  • SPF: an email setting that lists which servers may send mail for your domain.
  • TCPA: the US Telephone Consumer Protection Act, which governs robocalls and texts.
  • Teammate: in Vestra, an agent hired for a whole job, on the engine you pick.
  • Vault: in Vestra's plan, a store for an agent's card and logins, sealed, with limits a person sets. Coming later.
  • Wallet: a way for an agent to make payments within permissions you set, as in Manus Cue.

Related reading

Sources

All checked on 23 November 2026 on the makers' own pages unless stated.