Wake your assistant the moment a call ends

Two ways to find out a call is over

There are two ways for an assistant to learn that a phone call has ended. It can keep asking. Or it can be told.

Asking is the default. Your assistant looks at the line every so often, sees what's new, and acts on it. That works, in the way checking the letterbox works. But a call that finishes at 2:05 might not be noticed until the next look, and if it ended with "they can do Thursday, but only if you confirm today", the wait matters.

Being told is what we call Bot push. The line starts your assistant itself, the moment there's something to deal with. Nothing waits for a check-in.

What a webhook is

Bot push runs on a webhook, so here's the word in plain terms. A webhook is a private web address that belongs to one program, which another program can post a message to. Think of it as a doorbell. Your assistant gives the line its doorbell, and the line rings it when something happens.

With the address comes a key, which works like a password. The line sends the key with every message, so your assistant can tell a real wake-up from someone who's guessed the address. A message without the key isn't from us.

The three moments that can wake it

You choose which of these start your assistant. All three, or only the ones you care about.

• A call finishes. Your assistant gets the outcome as soon as the call is written up. If the call didn't get the job done, it can decide the next step without waiting to be asked.
• You text the line an instruction. "Move my Thursday cleaning to next week." Your assistant starts within seconds with that as its task, and you never opened it.
• Mail arrives. A quote lands at the line's email address. Your assistant wakes, reads it, and replies from the same address or leaves you a note.

Setting it up

It takes about two minutes and you do it once. Your assistant needs to be connected to the line already. If it isn't, connect the Bot you have first and come back.

• Make a routine that runs on a webhook. This happens in your assistant, not in CosVoice. In Grok, make a routine and choose the Webhook trigger.
• Copy the two values it shows. The address it posts to, and the key.
• Paste them into CosVoice. Open Settings, go to Connect, and find Wake your Bot. Both values go there.
• Send a test. Use Send test and check that your assistant started. If it did, you're done.

Or ask the assistant to do it

If copying and pasting between two windows isn't your idea of a good time, tell your connected assistant: "set up your CosVoice Bot push". An assistant that can make its own webhook routine, which a Grok Bot can, does the whole thing itself and tells you when it's finished.

If you added our ready-made Bot from the Bot page, it offers to set this up after your first call. Say yes.

What your assistant is handed

The message that wakes it is short. It says what happened in plain English and what to do about it, and it carries the ids the assistant needs to go and read the whole thing: the call, the text or the email.

That's on purpose. The message is a tap on the shoulder. The detail stays on the line, where your assistant reads it through the connection it already has. So the assistant wakes up, picks up the write-up and the transcript, does the work, and stops.

If the wake-up doesn't land

We try twice more, quickly, and then we stop. That might look stingy. But waking something late is no use: a nudge that turns up forty minutes on, in the middle of some other task, causes more confusion than it fixes.

Nothing is lost when that happens. The call is still written up and the instruction is still saved, and the mail is still on the line. Your assistant picks all of it up the next time it looks, and it can ask for everything since a given time, which is the thing to do after it's been offline. The written summary of every call comes to you by email as well.

Developers have a separate option. Signed webhooks are built for feeding call events into your own systems, and they retry over several hours. Signed webhooks for call events explains when you'd want those.

Do you need this at all?

If you use the line on its own, no. You call it, it calls for you, and the summaries arrive by email. There's no assistant to wake.

If you've connected an assistant and mostly ask for calls while you're sitting in the chat, it's nice to have. You'd have seen the result soon enough.

It's close to essential if you want things to happen while you're not looking. You text an instruction from a parking lot and the call gets made. A supplier rings back at six and your assistant has worked out the next step before you've read the summary. That's the difference between an assistant that does what it's told and one that follows up.

Turning it off again

You can switch Bot push off any time, or narrow it to one kind of event. If your assistant is being woken by every email and most of them are receipts, keep calls and instructions and drop mail.

Treat the key like any password. Keep it out of shared chats, and if you think it has got loose, make a new one in your assistant and paste that in where the old one was. The docs have the short technical version.

Bot push works with any assistant that has a webhook address and a token. So if you change assistants next year, the steps are the same: a new doorbell, pasted into the same place.

Common questions

More on setup guides

• Conditional call forwarding, explained without the jargon
• Set up an AI phone agent in ten minutes
• Connect Claude to a phone line
• Get ChatGPT to make a phone call