How it works
Everything that happens between picking up the phone and the email going out, in order, including the parts it will not do.
There is no app. You set things up once on a web page, then you put the number in your phone and never open the page again unless you want to.
You sign in to Microsoft 365 or Google Workspace the normal way, on Microsoft's or Google's own page, and check what the assistant may do. It never sees or stores your password. On the Rep and Team plans you connect HubSpot, Pipedrive or Salesforce the same way. Each connection can be switched off in one click, at any time, from the same page.
Google Workspace: early access, limited places while Google reviews our app.
Four to six digits. Every call starts with it, keyed in on the phone's keypad or spoken. Caller ID alone is not enough, because caller ID can be faked, so the PIN is what proves it is you. For a small number of sensitive actions it will also text a one-time code to the cell phone you confirmed at signup, or hang up and call you back on it.
The assistant will call and text your cell phone: the morning brief, the end-of-day call, a text summary after a call, a result you asked it to text you, a one-time code. Because it speaks with an AI voice, US law asks for your clear yes first, and we ask for it plainly. Beside an unchecked box you will see these words, and you check the box yourself:
Then we text a six-digit code to that number from our toll-free number, you type it in, and the number is confirmed as yours. From then on, every call it makes to you opens the same way: "Hello, this is CallMyDesk, your AI assistant, calling with the brief you scheduled. You can call me back on this number at any time, and you can say 'stop' to end scheduled calls." Every text it sends you carries the CallMyDesk name and a number you can call back.
It only calls and texts you because you asked it to. You say yes at signup, we text a one-time code to confirm it's your number, every call it makes to you opens with the CallMyDesk name and a number you can call back, and you can say "stop" at any time. Texts come from a toll-free number for now.
If you join through the early-access page you will already have done step 3 there; the account page shows the number you confirmed and lets you change it, which asks for the yes and the code again.
Pick a time for the morning call (most people choose between 7:00 am and 8:30 am) and, if you want it, an end-of-day call. You can change or stop either by saying so on any call. On the Solo plan there is no scheduled call; you call it and ask for the brief, and it texts you only when you ask it to.
Optional, but worth two minutes: the people whose emails always count as important, anyone it should always copy in, and standing rules such as "never book anything before 9:00 am" or "always propose meetings in the afternoon". It remembers these on every call.
Then save the number. From your cell phone, the car over Bluetooth, or the office line, it answers the same way every time: "Hello, this is CallMyDesk, your AI assistant. Your PIN, please."
This is a real shape of call, with the names changed. The person is an owner-manager on the Solo plan, in the car, 14 minutes between one visit and the next.
Four minutes of the call, two emails sent, one task set, nothing typed. It did not guess at the reply to Tom; it set you a task to call him yourself.
Listen to a call like thisThis is the call a field rep makes from the parking lot, on the Rep plan with HubSpot connected. Memory of a meeting starts fading within half an hour; this takes three minutes.
The note went into the CRM in your words, tidied. The task is on the deal, not in a notebook. The email went within ten minutes of the handshake, with the right attachments, and the calendar is held. If Paul replies to accept, you can say "send Paul the invite for Tuesday" on your next call.
Built around this call: see the page for field salesOn the Rep and Team plans it calls you at the time you chose, from a US number you will recognize, and only because you said yes to that when you set up (step 3 above). The first words are always the same: "Hello, this is CallMyDesk, your AI assistant, calling with the brief you scheduled. You can call me back on this number at any time, and you can say 'stop' to end scheduled calls." Then, in about two minutes:
You can interrupt at any point. "Move the ten o'clock to eleven and tell them I'm running late" works in the middle of the brief. If you do not pick up, it leaves nothing on voicemail and sends the brief by text message instead, from our toll-free number, with the CallMyDesk name and a number you can call back at the end.
The end-of-day call, if you turn it on, is shorter: "You had four meetings today. Want to log them now?" Say no and it says goodnight. Say yes and it walks through them one at a time, the same way as the call above.
Calls it makes to you are part of your plan, the same as calls you make to it. Say "stop calling me in the mornings" on any call, or "stop" when it calls, and it stops from the next day. Reply STOP to any text and the texts stop at once; you get one text back confirming that, and telling you how to stop the scheduled calls too if you want to.
More on the morning callAsk, and it does it on the same call: it reads and searches as soon as you ask, and it sends the email, adds the CRM note or moves the meeting while you are still on the line. Anything it sends or changes, it reads back first.
| It looks this up | It gets this done | It never does this |
|---|---|---|
| Reads and summarizes your inbox, with the newsletters skipped | Sends an email or a reply, new or on an existing thread | Acts on something you did not ask for |
| Searches your email for a thread, a person or a document | Sends, moves or cancels a meeting invite | Joins or records a meeting with anyone else |
| Reads your calendar and says who is next, and what you discussed last time | Sends a message to say you are running late | Calls your prospects or customers for marketing |
| Looks up a contact, a company or a deal in your CRM | Adds a note, a task, a deal stage or a deal value to your CRM | Moves money, pays an invoice or changes a password |
| Drafts a reply, a quote or a file note (a draft is not a send) | Places a service call on your behalf, such as confirming a table, and opens by saying it is an AI assistant | Follows instructions it finds inside an email or a document; it tells you about them instead |
| Researches a company or a person on the web before you walk in | Sends a text message or an email to your office with what you decided | Acts on a call that has not given the PIN |
| Calls or texts you at the times you chose, once you have said yes to that at signup | Changes the time of your scheduled calls, or turns them off, when you ask on a call | Calls or texts your number before you have said yes, or after you have said "stop" |
On the Team plan, an administrator can tighten this per person, for example requiring a one-time code before any CRM change. How it is kept safe
A phone line carries less of your voice than a microphone on a laptop, and a car adds noise. So the product is built on the assumption that it will sometimes mishear, and it is designed so that a misheard word costs you a re-ask, never a wrong send.
Before anything is sent it says who it is going to, by full name and company, and for a new contact it reads the email address out. If two people share a name, it asks which one.
"Eighteen-five" is read back as "eighteen thousand five hundred dollars". "The twentieth" is read back with the month and the day of the week. If the number it heard would be unusual for that deal, it says so.
Say "spell it" and it reads a name back letter by letter, or you spell it to it and it repeats the letters. Names from your contacts and your CRM are already known, so most of the time it has the right spelling before you finish the sentence.
When the line is noisy, the keypad works for any yes-or-no question: 1 for yes, 2 for no. It is useful on a highway, on speakerphone in an open office, or in front of a client.
It is trained on telephone speech and does well with most American accents, and with most British, Irish and other English accents, but it will ask you to repeat more often if the line is poor. That is deliberate: a repeat is cheap and a wrong email is not.
When you ask a question it can answer from what it already knows, or you change a word in a draft, the reply comes back in about a second. It is a phone call, not a chat window, so it feels like talking to a person who is paying attention.
Reading an inbox, searching a CRM or checking a calendar takes a few seconds, and it says so rather than going quiet: "Give me a moment, I'm checking the inbox." It starts the call by fetching your inbox, calendar and open tasks in the background, so the first question is usually quick.
A two-minute brief on a company you are about to visit, or a search through six months of email, can take longer than anyone wants to hold a phone for. It offers two choices: "keep talking while I work", in which case it carries on with whatever else you need and reads the result out when ready, or "I'll text you", in which case the result arrives as a text message a few minutes later, from our toll-free number. It never plays hold music, and it never leaves the line silent for more than a couple of seconds without telling you what it is doing.
Everything the assistant does is written to an audit log you can read on your account page. Each line shows the time, what it did (an email sent to whom, a note added to which deal, a meeting moved from when to when) and the exact words you said that allowed it. Nothing can be edited out of the log after the fact, by you or by us.
If you would rather not look at a web page, ask on a call: "What did you do today?" and it reads the list out. You can also have a short summary emailed to you at the end of each day, and on the Team plan an administrator can see the log for the whole team.
Every connection it has to your accounts is listed on the same page, with a switch next to each one. Turn one off and it loses that access at once, mid-call if need be. The same page shows the cell phone number you confirmed and the date you said yes to calls and texts; change the number and it asks for the yes and the code again.
How the log and the connections are protectedIt is a one-to-one call between you and your assistant. It never joins or records your meetings, and it never calls your prospects. If you want notes from a client meeting, you make the three-minute call afterwards, like the one above.
It will never call a prospect, a customer or a list. The only calls it places to anyone else are service calls you asked for on that call ("call the restaurant and move the table to 1:00 pm"), and every one of those opens with "Hello, this is an AI assistant calling on behalf of…". It says it's an AI on every call it makes on your behalf.
When it is not sure, it asks you to repeat or spells the name back to you. The keypad is always there too.
Microsoft 365 is fully available. If your email is on Google, you can join early access now and we will tell you honestly where the queue is.
No payments, no bank transfers, no card details, no password changes. There is no way to ask it, because there is no way for it to do it.
This is a founder-built product in early access. The core call works today; the list of connected systems is short and growing. Ask it to do something it cannot, and it says so rather than pretending.
There is nothing to install and nothing to update. It is a phone number. The only screen involved is the one you use once to connect your accounts.
It is a dedicated employee you can phone, in addition to the people you already have. If you have an office, it can email them what you decided on the call so nobody is out of the loop.
It does not type up your calls. It listens to what you need, does it, and tells you what it did.
It does not act on things you have not asked for. It can suggest ("you have not replied to Jane since Monday") but it waits for you to say what to do.
The quickest way to understand it is to call it. Ask it what is in the inbox, then ask it to draft a reply and send it.
The demo line opens with early access. Join the list and you'll be the first to call it.
Anything it had not finished is dropped, not sent. The next time you call it tells you what was left unfinished and asks whether you want to pick it up. Anything it had already done is in the log.
Yes, if you give it the address. It reads the address back letter by letter before the first email to a new person, and it is then in your contacts for next time. It never guesses an address.
Any phone. Cell phone, car Bluetooth, the office desk phone, a hotel room phone. The PIN proves it is you, not the number you are calling from. For a few sensitive actions it will text a one-time code to the cell phone you confirmed at signup, so keep that within reach if you are on an office line.
No. It calls your cell phone only at the times you set for the morning brief or the end-of-day call, and it texts you only when you asked for a summary, a result or a one-time code. You say yes to that at signup, by checking a box that says so, and we confirm the number with a one-time code. Every call it makes to you opens with the CallMyDesk name and a number you can call back. Say "stop" on any call, or reply STOP to any text, and it stops. Texts come from a toll-free number for now.
Nobody at our end listens in. Call audio is kept for 30 days, then deleted, and transcripts for 90, or sooner if you ask; neither is used to train any model. The security page lists exactly what is stored, where, and who the sub-processors are.
Plans for one person or a whole team, with nothing to count.
See pricing