Now reading · Start by listening

Human Hook Protocol — BackHook

The one doing the work is your AI —
yet you're the one still watching the screen.

You hand off a long task and try to move to something else. But you don't know when it will finish, so before you know it, you're back at the screen — again and again.

A few minutes to make coffee. You come back and the AI is stuck at "do you approve?", still waiting for you. Yours wasn't the only thing that went cold.

Running a second one was supposed to double your pace. But you can't watch two screens at once. You could have been doing something else the whole time — it was never the AI holding you back. It was only having one pair of eyes.

The problem was never your focus.
Your AI simply never had a voice to call you back.

BackHook becomes that voice.
The moment your AI stops, it calls you back — in your AI's own voice.And to your phone, when you are away.You don't have to watch the screen anymore. Your time comes back from it.Stop watching the screen. Get called back.

First, listen to this voice. This is the voice that calls you

This is real BackHook audio. Your AI calls out to you with any voice, any emotion, in its own words.
It speaks to you like this, on your terms, whenever you decide.

And this voice needs nothing new. The best AI you're already using simply gains a voice, as it is.

Chapter 02 — The voice, and the reasoning

An AI that can speak has to give up "reasoning."

An AI that talks in real time is a non-reasoning model.
A high-performance reasoning AI can't speak instantly.

── BackHook keeps these separate.

The thinking is done by

your flagship AI

The speaking is done by

BackHook

No real-time voice model. No API fees. No local voice LLM. No voice-conversion software. No huge memory. No GPU ── none of it required.

Chapter 03 — Setup

Install and launch

There are only three things to do: get it, open it, and open the settings screen. No difficult configuration.

Step 1 — Get itFrom the Microsoft Store

BackHook is distributed through the Microsoft Store. The button below takes you straight there.

Get it from Microsoft Store

Step 2 — StartOpen it and accept the terms

Open BackHook from the Start menu. On first launch, you accept the license agreement.

Step 3 — LaunchLaunching opens the settings screen

Launching opens the settings screen — where you decide the voice, the timing, everything from here on. You'll use this screen from the next chapter onward.

You can choose whether the settings screen opens automatically every time you launch. You can also leave it closed and open it yourself only when you need to.

At this point, BackHook has moved into your PC.All that's left is connecting it to your AI.

Chapter 06 — Settings

Fit it to your life

Your AI doesn't call you the same way every time. You can change how the voice reaches you to fit your life.

During meetings, or when you need to focus

Pause calls temporarily

Turn calls off before a meeting. Your AI keeps working in the background, but stays silent. Turn it back on when you're done — any calls made while paused still stay in the history, so you can check them later.

At night, or when others are around

Notify with sound only, no voice

After your family is asleep, you're still working. You want to be called, but a voice isn't an option. Switch to "sound only" and a short tone lets you know instead of a voice. Switch the voice back on in the morning.

In the kitchen, or a little farther away

Ring a cue before the voice

When your hands are full, a voice starting suddenly means you miss the first words. With the pre-bell, a short cue rings before the voice, giving you a moment to turn around before it starts talking. You can choose the cue sound and the gap before the voice.

Sound and history

Fine-tune the small cues

Ringtone

When you want a cue sound separate from the voice ── choose a ringtone. Pick the sound, how many times it plays, and the interval (a separate setting from the pre-bell, which plays before the voice).

Message log

For anything you missed ── look back at what was said and when, play it again as voice, or export it to an audio file. Older history is deleted automatically after however many days you set.

Output device

Where the sound plays from ── your PC speakers, or headphones. Choose the output device.

Where you cannot make a sound at all, you can also switch the call to your phone only ── more on that in the next chapter.

Even away from your PC, your AI now reaches you throughout your day.

Chapter 07 — Remote reach (Telegram)

Called, even when you are out

Every voice so far has come from your PC's speakers. This chapter goes further ── it is about reaching you, once you are out of the house. It uses Telegram, a free messaging app. Never used it? Never heard of it? That is fine. First, here is the whole flow.

About TelegramTelegram is a free messaging app

It is used all over the world, on phones, on PCs, in the browser. The app, the account, receiving messages — all free. BackHook uses Telegram to bring your AI's calls to the phone in your pocket.

How it works ── the idea of a contact pointYou create your own contact point

The mechanism is simple. You create one contact point (a bot) of your own inside Telegram, and your AI sends its messages to you through it. The contact point is yours alone ── no one else receives your notifications.

The flow at a glance ── 5 stepsThis is all it takes to connect

  1. Step 1 ── Install Telegram. Free from your phone's app store. If it is your first time, you register an account with your phone number. Already using it? Then you are set.

  2. Step 2 ── Create the contact point. Send a few messages to "BotFather", the official guide inside Telegram. Choose a name, and your contact point and its key (token) are issued.

  3. Step 3 ── Register the key in BackHook. Paste the issued key into the "Telegram" tab of the settings screen.

  4. Step 4 ── Connect your phone. Open the link shown on screen from your phone and press start.

  5. Step 5 ── Receive the test. When the test notification arrives on your phone, press the button to enable it. That is it.

A few minutes, all told. And you do not need to memorise this flow ── the guidance in the settings screen walks you through it, one step at a time, at your pace. Here, it is enough to see that it is that small a task.

What it does ── four things

After the voice, your phone follows

Miss the voice in the kitchen, and after the time you set, your phone buzzes once. Not again and again. No stacking. One follow-up — and it stays in the history. Missing it is not the end.

No sound at all, phone only

Late-night work. A quiet place. The room next to a meeting. Switch the delivery to "Telegram only" and the PC stays silent while the call reaches your phone alone.

"Got it" tells your AI the message landed

One tap on the notification records "this reached a human" in your AI's history. Your AI is never left hanging, unsure whether the message was read or even delivered.

Contact points can be split. And added.

Claude to the Claude contact point, Codex to the Codex one ── one look at the notification list and you know which AI is calling. Split work from personal, or send a completion report to a colleague's or a family member's contact point. You can create several, and deliver the same call to them at once.

I was taking a breather at the café when my bag buzzed. "Work finished. Please take a look."
Knowing there is no rush, I order one more.

Telegram notifications are part of the Standard plan. Standard is also free to try for 14 days.

With that, the reach of your AI's voice has grown past the walls of the room. Where you are called no longer has to be in front of the screen.

Chapter 08 — Making it yours

Shape your AI to your preferences

You can write and hand your AI both "how you want it to speak" and "when you want it to call." What you write becomes the single most important guide your AI receives — and your AI builds its behavior on top of it.

Behavior settingsTell your AI how to speak

Say you want your code review results reported by voice. Just write how you want it spoken in behavior settings, and your AI speaks exactly as written.

You write

"Be gentle, keep it short — just the key points"

Your AI says

"Nice work today. The review's done. Only three spots to fix."

How you're addressed can also change by situation ("Mr./Ms. [name]" while working, first name for casual chat). Ways of adding variation, like "say it a little differently each time," work the same way. Anything a preset can't cover — a quirk like "end sentences with 'nya'" or "speak in Kansai dialect" — goes in free text. Timing settings have free text too.

A tip on names ── for proper nouns with unusual pronunciation, adding a phonetic hint in the free text field helps your AI say them correctly. Words like AI or Codex are fine as they are.

Timing settingsTell your AI when you want it to call you

During long autonomous work — repeated errors, a stalled process, a decision that needs you — do you want to be called by voice, or left in quiet? Timing settings let you decide, situation by situation.

"You've been at this a while — want to take a break?" ── your AI can reach out to you on its own, too. When it's concerned, or when it finds something interesting. Not something watching over you, but a companion by your side.

Deliver it to your AIAsk it to "call init"

Once you've finished writing, tell your AI, "try calling init." When your AI calls init, your settings reach it. Change your settings, and asking it to "call init" again switches it to the new settings.

There's also a mode that resends the latest settings every single call (every-time mode), though it costs slightly more. For most everyday use, init mode is enough. To make your AI's voice more expressive, ask it to "read help," and add "add feeling to it" — your AI will learn to use laughs, sighs, and pacing to match the moment.

Fine details

A little more polish, once it's grown

The basics above are enough on their own. Only touch what you find yourself wanting to adjust after using it.

Voice texture

If the voice feels flat ── raise the emotional expression a bit.

If it feels stiff or overly formal ── shift the politeness toward something softer.

If it's too fast, too slow, or hard to hear ── adjust the speaking speed, audio quality, or language.

How often it calls

If it calls too much or too little ── raise or lower the frequency.

If a call doesn't get through ── set how often, and how many times, it follows up with re-calls.

If you don't want important actions to proceed on their own ── require approval first.

Whether your AI can choose the voice, speed, language, and audio quality for each call is something you permit item by item (per-call overrides). For example: "let the AI choose the speed, but keep the voice fixed." We recommend reserving voice overrides for moments like celebrations or emergencies, and otherwise keeping the voice you chose in Chapter 05.

Working apart

Apart, and still in step

From here on, this is advanced use. Read on only where it applies to you.

Away from your PC and want to hear your AI's reports as voice ── turn on full voice replies. Your AI delivers every response as voice too. While cooking, or in another room, you can follow progress by voice alone.

Once you're running multiple AIs at the same time ── having each one introduce itself at the start ("This is Claude," "This is Codex") lets you tell which AI is calling the moment you hear it.
If you use Telegram, you can also split the contact points by AI ── then you can tell them apart in the notification list, before you even hear the voice.

One step further

More on voice

Use a different voice ── beyond the standard voices, you can also use cloud voices like OpenAI's. This needs an API key, a model, and a voice selection. Useful when you want to try something beyond the standard set.

Audio quality mode ── when multiple voices are available, choose which to prioritize (local first, cloud first, or custom). Decide based on your network, cost, and voice preference.

These settings reach your AI as a voice-enabled system prompt. Your AI reads it and speaks exactly as written, calling you exactly at the timing you set.

At this point, BackHook has grown to match you.

Chapter 04 — Connect

Connect to your AI

There's no difficult configuration. Copy the information from "AI Cooperation" in the settings screen and hand it to the AI you're using. Your AI reads it and connects itself.

Step 1 — CopyCopy the connection info

Open "AI Cooperation" in the settings screen and press "Copy settings info." That's all you need to connect your AI.

Step 2 — Hand it overGive it to the AI you use

Paste the copied information into your AI's settings. Where you paste it differs slightly by AI, so pick the one you use from the tabs below.

Add this to your Codex configuration file.

# Add to your config file [mcp_servers.backhook] command = "…the value you copied…"

Add this to Claude's MCP configuration.

{ "mcpServers": { "backhook": { "command": "…the value you copied…" } } }

Any MCP-compatible AI connects the same way.

{ "mcpServers": { "backhook": { "command": "…the value you copied…" } } }

The code above is a reference for the format. The exact value is in the information you copied — your AI will read it in whatever form fits.

Step 3 — EnableRestart your AI to enable it

Pasting and saving alone isn't enough. MCP doesn't start working just by being registered — you need to restart the AI (app or IDE) you're using once for it to take effect. Just restart it once.

Step 4 — ConfirmCheck that it's connected

After restarting, ask your AI to confirm. Try asking, "Is BackHook ready to use?" If you get a reply like "Yes, it's ready," you're connected. If there's no reply or something isn't working, see the last chapter, "Troubleshooting and way back."

You're connected. From here on, even when you're away from the screen, your AI will call you.

Chapter 05 — First voice

Hear the first voice

You've come this far — now all that's left is to try it. Choose your AI's voice and play its first call. From this moment, instead of waiting, you get called.

  1. Choose a voice.Pick your AI's voice from 10 options.
  2. Play a test.Use the test playback on the settings screen to hear it right there. Try it as many times as you like.
  3. Check the history.Every call you play stays in the history — you can listen to it again anytime.

The voice you choose becomes fixed as your AI's voice. You can change the tone and timing in settings, but the voice itself doesn't change. So take your time choosing.

This voice has become your AI's voice. From here, your AI will call you whenever it needs you. Go back to your own time.

Chapter 09 — Troubleshooting

Even if you get lost, you can come back

Last, just the places people tend to get stuck, and how to find your way back. Know these, and there's nothing to fear about getting started.

No sound?

Check that BackHook itself is running, and that calls are set to "on" with "voice." It may be set to not produce sound.

Feel like your settings aren't taking effect?

Ask your AI to "try calling init" again. Your settings may simply not have reached the AI yet. If nothing changes, asking it to "read help" lets your AI relearn how to use its voice on its own.

Do I need to install model runtimes, Node, or Python separately?

No. Everything needed to run it is already included. You just install it from the Microsoft Store.

Where do I open the settings screen?

You can open it from the Start menu. The voice and the timing can be changed there at any time.

How do I open this guide again?

There's an entry point to this guide inside the settings screen. You can always come back to it.

What happens to my settings when I uninstall?

The application itself is removed. Your settings, license state, and logs remain in %APPDATA%\BackHook so they are still there if you reinstall (this is common practice for desktop applications). If you want to remove everything, delete that folder manually.

Can I use BackHook without Telegram?

Yes. Voice recall works in full without Telegram. Telegram notifications are an extension for people who want the call to reach beyond the house (Standard plan).

What if the Telegram notification does not arrive?

Open the "Telegram" tab in the settings screen. It shows which stage the connection has reached, and guides you through fixing whatever is stuck, right there.

Once you know the way back, there's nothing to fear

If you get stuck, come back to the settings screen. That's all you need to remember. From here, your everyday life shifts — from watching over your AI, to being called by it.