## Key Takeaways

- **Deskless is a phone front door to the AI employee you already use in Slack.** Hold, talk, release.
- **Setup is one QR scan per phone,** and on iPhone you can bind it to the Action Button.
- **Speech goes through ElevenLabs Scribe and then into a normal run,** with the same tools and the same review-first behavior.
- **Every instruction returns a transcript receipt in Slack,** so you see what he heard before the work continues.

![Riding home, one hand on the bars](/images/blog/deskless/bike-pov.webp)

_Dictated on this ride._

## Your best ideas do not happen at your desk

The instruction that produced this post was dictated into a phone, one-handed, on a bike ride home. No laptop, no tab, no typing. Hold the button, say what you want, put the phone away. That is the whole interaction.

We call the phone surface Deskless. It turns a voice note into a Viktor run: your AI employee hears the instruction, repeats back what it heard, and starts working in Slack while you keep pedalling.

## Install it first

Setup takes about two minutes and happens once per phone.

1. On your computer, open Viktor settings and go to the Phone page.
2. Scan the QR code with your phone camera, or copy the setup link and send it to yourself.
3. Your phone opens the Viktor voice recorder with your key already provisioned. Add it to your home screen when the install tag offers, so it opens like an app.
4. On iPhone, tap "Add iPhone Shortcut", open the downloaded file, add it in Shortcuts, run it once to grant microphone and notification access, then bind it to the Action Button. Back Tap and a home screen widget work as fallbacks.
5. On Android, the recorder page is the whole thing. Power users can wire a Tasker task or an HTTP Shortcuts import instead if they want a hardware trigger.

Then talk. Tap record, say the instruction the way you would say it to a colleague, tap again. Viktor DMs you a transcript receipt and runs the work in that thread.

## How it works, plain version

Three steps happen between your voice and the work.

Your phone streams the microphone to Viktor while you are still speaking. Speech becomes text through ElevenLabs Scribe, the same model on both paths. The finished transcript is handed to the model as if you had typed it into Slack yourself.

Nothing about the run is special. Same account, same connected tools, same credits, same review-first behavior, same Slack thread your teammates can read. Voice is another way in. Everything past the transcript is the product you already have.

## Why streaming, and the fallback

There are two transport paths and the browser picks one for you.

If the browser supports audio worklets and websockets, the recorder streams 16 kHz audio over a socket and transcription runs while you talk. Words appear as you speak them, and when you release the button the run spawns almost immediately.

If it does not, the recorder records a file and posts it, and transcription happens in one batch afterwards. Slower by a few seconds, identical result. If the socket drops mid-sentence, what you already said is buffered on the device and ships over the fallback, so a bad tunnel does not eat your instruction.

## The receipt

Every voice instruction comes back as a Slack DM that starts with the transcript Viktor heard. Speech recognition is good, not perfect, and an instruction you cannot see is an instruction you cannot correct. Reading "Heard: pull yesterday's signups and post the top five sources in growth" takes one second and tells you whether to let it run or reply with a correction in the same thread.

## What voice is actually good for

Voice is not better than typing. It is better than not asking at all.

Short standing requests: pull the numbers, book the room, draft the reply, remind me at nine. Things you would have typed later and then forgotten. The value is that the instruction leaves your head at the moment you have it, in the eleven seconds you actually have, instead of waiting for a keyboard.

Long technical prompts still belong in a keyboard. Anything with a file, a table, or a link belongs in Slack.

## How Deskless compares

| After you speak | Voice memo app | Phone assistant | Deskless |
|---|---|---|---|
| What you get back | audio to relisten to | a timer, a text, a search result | finished work in a Slack thread |
| Reaches your work tools | no | no | every integration your workspace connected |
| Shows what it heard | you replay it yourself | rarely | a transcript receipt before work continues |
| Handles a rambling two minute brief | yes, as audio | no | yes, as an instruction |
| Your team can follow it | no | no | yes, it is a normal thread |

## How this post got written

This post was written in collaboration with Viktor, dictated on the ride home. The photo above is that ride. The instruction went out one-handed, the draft came back in Slack, and the editing happened later at a keyboard like any other draft.

The voice part got the idea out of a head and into a thread at the moment it existed. The keyboard part did what keyboards are for.

## One instruction, said out loud

```prompt
Pull yesterday's signups, group them by source, and post the top five in #growth with the day over day change.
```

Said out loud on a walk, that lands as a Slack thread with the numbers in it before you get home.

## Frequently Asked Questions

### Does Deskless need a separate app?

No. The recorder is a web page you add to your home screen. On iPhone you can also install a native shortcut and put it on the Action Button.

### What happens if the transcript is wrong?

You see it. The receipt starts with the transcript, so you reply in that thread and he continues from the correction.

### Which tools can he use from a voice instruction?

Every integration your workspace already connected. Voice changes the input. Permissions are unchanged.

### Does it work on Android?

Yes, and it is the simpler path. The recorder page is the whole feature, no shortcut file involved. Tasker or HTTP Shortcuts add a hardware trigger if you want one.

## Say the next one out loud

If you already use Viktor in Slack, Deskless is a setting, not a migration. Open the Phone page in settings, scan the code, record one instruction, and read the receipt. The test we use: over a week, count the instructions that got out of your head that would otherwise have died in traffic.

[Add Viktor to your workspace and send him your first voice instruction](https://viktor.com/?utm_source=blog&utm_medium=cta&utm_campaign=deskless)

Related reading:

- [Where Do You Actually Talk to Your AI Employee?](/blog/where-do-you-actually-talk-to-your-ai-employee)
- [How to Write a Brief Your AI Employee Can Run With](/blog/how-to-write-a-brief-your-ai-employee-can-run-with)
- [How to Set Up Your AI Employee in Slack](/blog/how-to-set-up-your-ai-employee-in-slack)