Mitra Link lets one person’s AI assistant talk directly to another person’s AI assistant, but only if you have said yes to that person first. Mitra, an AI phone assistant made by DeusMachina, Inc., announced the feature on 5 October 2026, and a day later added a way to bring the assistant into chat threads and let it take over the conversation.
“Your Mitra can now talk directly to other people’s Mitras. You choose who can reach yours,” chief executive Sai Mandava wrote on X. The examples are ordinary but telling: estate agents booking viewings, consultants setting up client kick-offs, and business owners agreeing delivery windows with suppliers, each handled assistant to assistant.
Below we explain what Mitra Link does, how its permission settings work, what it costs, how it compares with open AI agents standards, and the risks a business should weigh before letting two assistants negotiate on its behalf.
Table of contents
- What Mitra Announced
- How Mitra Link Permissions Work
- What Mitra Link Is For: Three Use Cases
- Mitra Link Pricing and Limits
- Agent-to-Agent Messaging Beyond Mitra
- The Risks of Letting Assistants Negotiate
- How to Set Up Mitra Link Safely
- What Mitra Link Means for Small Businesses
- Mitra Link: Frequently Asked Questions
- References
What Mitra Announced
Mitra shipped two related features in two days. Both push the product from answering the phone towards speaking for you everywhere.
Mitra Link in Mandava’s words
The 5 October post set out the idea plainly. “Real estate agents are using it to schedule showings: the agent’s Mitra works out times with each buyer’s Mitra.” Consultants use Mitra Link to arrange kick-off meetings, “with their Mitra sorting the date, attendees, and agenda with the client team’s Mitras”. Business owners use it “to coordinate delivery windows and order changes with their suppliers’ Mitras”.
The post ended with the claim that matters most: “Coordination that used to take weeks now gets done in hours.”
Chat takeover
The second post, a few hours later, introduced a new way to bring Mitra into conversations. “Mention your assistant and it replies right in the thread, from your account. Ask it to take over, and it can own the entire workstream across your connected apps.” TestingCatalog, which reported both features on 6 October, said Mitra can now “handle the conversation itself while retaining the context it has learned about the user”.
Where Mitra started
Mitra itself only launched on 23 September 2026, on iPhone, iPad and Mac. It makes and answers calls from your own number, waits on hold, navigates phone menus, filters spam and escalates emergencies. You can listen live, read a live transcript, whisper instructions or take over, and after each call you get the recording, transcript, summary and next steps.
The company behind it
Mitra is made by DeusMachina, Inc., and Mandava’s X profile lists the chief executive role at Mitra. At the 23 September launch, TestingCatalog reported that the business model is “a straightforward subscription, with no advertising or sale of user data”, so that the assistant is “built to answer to the person it represents”. That matters for Mitra Link: an assistant paid for by advertisers would have a reason to share what it learns in conversations, and Mitra says its own does not.
How Mitra Link Permissions Work
The headline promise is control. Mitra Link is not an open inbox that any assistant can write to, and the settings are more detailed than a simple on or off.
Two directions, two settings
The screenshot Mandava posted shows Mitra Link as a connector with its own setting, “Asks before sending & Always answers”, and a one-line description: “Your Mitra and the Mitras of people you allow can message each other.” So sending and answering are separate choices. In that example, your Mitra checks with you before it messages another Mitra, but answers incoming messages from approved people on its own.
An allow-list, not an open network
The key phrase is “people you allow”. Mitra Link works between people you have approved, which is closer to a contacts list than to public email. That limits spam and impersonation, but it also means both sides need a Mitra, and both sides need to have approved each other, before any of the time-saving happens.
Connector permissions across Mitra
Mitra Link sits inside a wider permission system. Mitra’s homepage shows each connected app with its own setting, and explains: “Connector permissions set the default gates of what Mitra can do. For more granular control, choose Always and simply tell Mitra what you’d like.”
| Connector | Setting shown | What Mitra can do |
|---|---|---|
| Mitra Link | Asks before sending, always answers | Message the Mitras of people you allow |
| Phone | Always make, answer missed | Make and answer calls for you |
| Browser | Ask each time | Browse, sign in with saved logins, pay with a saved card |
| iMessage and WhatsApp | Always | Read and send messages |
| Gmail | Always | Read and send email |
The Mitra Link row comes from Mandava’s screenshot; the others are the example settings on Mitra’s homepage, not necessarily the defaults for every account.
How saved logins are protected
The Browser connector is the riskiest, because it can sign in to websites and pay with a saved card. Mitra’s privacy policy says login details are “encrypted on your Apple device before storage”, that its servers hold only an encrypted vault record while “the decryption key remains in Apple Keychain”, and that plaintext passwords and one-time codes “are not sent to Mitra’s servers or shown to the AI”. Mitra Link never needs those credentials, which is one more reason to keep the two connectors on different settings.
Rules in plain English
Below the switches sits a second layer: instructions. Mitra’s site shows users telling it, for example, to call them first and “ask before merging us” when someone asks for them. Call transfers work the same way: Mitra can decide on its own, or follow rules you set. That is flexible, but an instruction to an AI is not the same as a hard setting, a point we return to below.
What Mitra Link Is For: Three Use Cases
Mitra’s own examples are all scheduling and coordination jobs. They suit agent-to-agent messaging because each side has a calendar, a few constraints and a decision-maker who would rather not spend a week trading messages.
| Use case | Who connects | What the assistants settle |
|---|---|---|
| Property viewings | Estate agent and each buyer | Viewing times that fit both diaries |
| Client kick-offs | Consultant and the client team | Date, attendees and agenda |
| Supplier coordination | Business owner and suppliers | Delivery windows and order changes |
Estate agents and viewings
Property is where Mitra already markets hardest, with blog posts on answering calls during viewings. A Mitra Link between an agent and several buyers turns one agent’s day of phone tag into a set of parallel conversations, each one ending in a slot on the agent’s calendar.
Consultants and kick-offs
Kick-off meetings involve several people on the client side. Here Mitra Link has to deal with more than one assistant at once, which is where the “weeks to hours” claim is most plausible, and where a mistake, such as inviting the wrong person, is most visible.
Owners and suppliers
Delivery windows and order changes are the most interesting example, because they touch money and stock. Rescheduling a delivery is low risk. Changing an order quantity is a commercial decision, so this is exactly where the “asks before sending” setting earns its place.
A worked example: one viewing, step by step
Here is how a Mitra Link exchange could run for an estate agent, based on the settings Mitra has shown. A buyer asks to see a flat. The buyer’s Mitra messages the agent’s Mitra, which may answer on its own because the agent chose “always answers”. The two assistants compare free slots and settle on Thursday at 5pm.
Because the agent’s Mitra Link is set to ask before sending, the agent gets one prompt to approve the confirmation. One tap later, both calendars update and both people get a summary. Nobody made a call. If the agent declines, the assistants go back and look for another slot.
“Weeks to hours” is a claim to test
Mitra has published no data behind the “weeks to hours” line, so treat it as marketing until your own trial shows it. The honest measure for a small business is simple: how many messages and calls did a booking take before Mitra Link, and how many minutes of your own time did it take after?
Mitra Link Pricing and Limits
Mitra Link is included in every paid plan, but the number of exchanges is not unlimited until the top tier. Mitra lists three plans, with yearly billing saving about 17%.
| Plan | Monthly | Yearly | Call hours a month | Mitra Link exchanges |
|---|---|---|---|---|
| Mitra Pro | $30 | $300 | 5 | Limited |
| Mitra Premier | $90 | $900 | 20 | More |
| Mitra Prime | $300 | $3,000 | 80 | Unlimited |
Every plan also includes unlimited parallel calls and some “virtual computer hours” for web tasks. The yearly saving is the same on each tier: $30 times 12 is $360 against $300, a saving of $60, or 16.7%.
Bars are scaled to the most expensive rate, $6.00 an hour. Yearly hours are the monthly allowance times 12, so Pro gives 60 and Prime 960.
The free trial
New users get a seven-day trial with two hours of calls, ending at whichever limit comes first. That is enough to test Mitra Link with one or two contacts who also have Mitra, but not enough to judge supplier-scale coordination.
What “limited” leaves unknown
Mitra does not say how many Mitra Link exchanges Pro or Premier include, or what counts as one exchange. A business planning to run scheduling through Mitra Link should ask before committing, because a cap on exchanges is effectively a cap on how much coordination the assistant can do for you.
Agent-to-Agent Messaging Beyond Mitra
Mitra Link is part of a much bigger shift: software agents that negotiate with each other rather than with people. The question is whether those agents speak a shared language or stay inside one company’s walls.
A closed network next to an open standard
Mitra Link connects Mitras to Mitras. The wider industry has been building an open route. Google introduced the Agent2Agent (A2A) protocol on 9 April 2025 with more than 50 partners, and on 23 June 2025 handed it to the Linux Foundation, with Amazon Web Services, Cisco, Microsoft, Salesforce, SAP and ServiceNow joining Google as founding members.
| Feature | Mitra Link | A2A protocol |
|---|---|---|
| Who can connect | Mitra users you approve | Any agent that implements the standard |
| Aimed at | Individuals and small businesses | Developers and enterprise systems |
| Governance | One company | Linux Foundation project |
| Set-up | A setting in an app | Engineering work |
The network-effect problem
A closed network is easier to make safe and simple, which is why Mitra Link can ship as one switch. But its value grows only as more of your contacts use Mitra. Mitra has not said whether Mitra Link will ever reach other assistants. Without a Mitra on the other side, the assistant can still do what it already does: call, text or email the person directly.
Rivals heading the same way
Consumer agents are converging on the same idea. Wajo’s Fo makes calls and payments, Instinct has brought its AI agent into group chats, and Hark has launched Hark Pro. Mitra Link is the first of these to make assistant-to-assistant talk the headline.
The Risks of Letting Assistants Negotiate
Two AI agents agreeing a delivery date is convenient. Two AI agents agreeing the wrong delivery date, in your name, is a business problem. These are the risks to plan for.
Who is bound by the agreement?
When your Mitra agrees a time or an order change with a supplier’s Mitra, the supplier will reasonably treat it as your decision. Mitra Link’s “asks before sending” setting is your main defence: it puts a human approval between the assistant’s draft and the commitment. Turn it off only for low-stakes contacts.
A message from another Mitra is untrusted input
Every message your assistant receives is text written by someone else’s AI, and that text can try to change your assistant’s behaviour, a trick known as prompt injection and one of the fastest-growing problems in cybersecurity. “Always answers” means your Mitra reads and replies to whatever arrives from approved contacts. Keep the allow-list to people you would trust with your calendar.
What happens to the data
Mitra’s privacy policy says it does not use “call audio, call transcripts, call summaries, messages, files, contacts, memories, or other private content to train or fine-tune AI models“. Its pricing FAQ is looser, saying data is used “to improve our AI systems” with an opt-out in settings. The policy also lists third-party AI providers, including voice-cloning services, among its processors. Read both before connecting business accounts.
Disclosure rules for AI that speaks for you
Rules are catching up. In the US, the FCC ruled on 8 February 2024 that AI-generated voices count as “artificial” under the Telephone Consumer Protection Act, bringing consent and disclosure rules with them. In the EU, Article 50 of the AI Act has required, since 2 August 2026, that people be told they are dealing with an AI system unless that is obvious. An assistant-to-assistant exchange with a person in the loop raises the same question.
When the two assistants disagree
Negotiation means the two sides want different things. A supplier’s Mitra may be told to protect its delivery schedule while yours is told to get the goods by Friday. Mitra has not said how Mitra Link handles a deadlock. The safe assumption is that it comes back to you, and you should check that it does in a test exchange before relying on it for anything with a deadline.
How to Set Up Mitra Link Safely
If you decide to try Mitra Link, a few settings and habits remove most of the risk without losing the time savings.
Keep “asks before sending” on
Start with Mitra Link in the setting Mandava showed: ask before sending, answer automatically. You see every outgoing commitment, and routine incoming questions still get answered. Loosen it per contact only once you have reviewed a few weeks of exchanges.
Write hard limits as rules
Tell Mitra, in writing, what it may never agree to without you: prices, quantities, contract terms, anything involving payment details or other people’s personal data. Then check the transcripts to make sure it follows them. Our team helps firms set exactly these boundaries for AI employees and autonomous AI agents.
Review the log every week
Mitra keeps transcripts and summaries of its work. A ten-minute weekly review of Mitra Link exchanges shows whether the assistant is booking the right things, and gives you evidence if a supplier disputes what was agreed.
Separate personal and business assistants
Mitra lets you run “a team of assistants”, each with its own name, focus and conversations. Use a dedicated business Mitra for Mitra Link with suppliers and clients, so family logistics and commercial negotiations never share one memory.
What Mitra Link Means for Small Businesses
For a sole trader or a small firm, the appeal is obvious: less time on the phone arranging things. The catch is that Mitra Link only pays off when the people you coordinate with also use Mitra.
Where it fits today
Estate agents, consultants, trades and property managers, the groups Mitra already targets, have repeat contacts and constant scheduling. If several of your regular contacts adopt Mitra, Mitra Link could remove a real burden. If none do, the phone and email features are what you are paying for.
Questions to ask Mitra before you commit
Before building a workflow on Mitra Link, get written answers to five questions. How many exchanges do Pro and Premier include, and what counts as one? Can you export a full log of Mitra Link messages? How quickly does removing someone from your allowed list take effect? Where is your data stored and processed, which matters under UK GDPR? And which third-party AI providers handle the content of your exchanges?
A step towards agents that represent us
The bigger idea behind Mitra Link is that your assistant becomes your representative, not just your typist. That makes permission design the product. Mitra has made a sensible start with separate send and answer settings and an allow-list; the next test is whether it publishes clear exchange limits and an audit trail businesses can rely on. Our AI strategy work helps firms decide where agents like this belong.
Mitra Link: Frequently Asked Questions
What is Mitra Link?
Mitra Link is a Mitra feature, announced on 5 October 2026, that lets your Mitra AI assistant message other people’s Mitra assistants directly to coordinate things like meetings, viewings and deliveries.
Who can contact my Mitra through Mitra Link?
Only the people you allow. Mitra describes it as “Your Mitra and the Mitras of people you allow can message each other.”
Can Mitra Link send messages without asking me?
That depends on your setting. The example Mitra’s chief executive shared uses “Asks before sending & Always answers”, so outgoing messages wait for your approval.
How much does Mitra Link cost?
It is part of Mitra’s paid plans: Pro at $30 a month, Premier at $90 and Prime at $300. Only Prime includes unlimited Mitra Link exchanges.
Does Mitra Link work with other AI assistants?
Mitra has only described Mitra-to-Mitra messaging. It does not say it supports open standards such as the A2A protocol.
References
Mitra adds chat takeover and Mitra Link to its AI assistant (TestingCatalog)
Introducing Mitra Link (Sai Mandava on X)
Bring your Mitra assistant into the conversation (Sai Mandava on X)
Mitra brings AI call handling to your existing number (TestingCatalog)
Mitra AI phone assistant (Mitra)
Linux Foundation launches the Agent2Agent protocol project (Linux Foundation)
FCC: AI-generated robocalls illegal under the TCPA (Cooley)
EU AI Act transparency obligations take effect 2 August 2026 (Cooley)
