---
title: "Devices"
description: "Work on a folder on another machine — an edge box, a server, a lab machine — from your own computer."
---

## What a device is

A device is another machine of yours with Ysra installed and signed in: an edge
box with the GPU and camera, a build server, a lab machine.

Once it's a device, a folder on it works like a folder on your laptop. I read
it, run its checks, and change it when you ask, **on that machine**, with its
real hardware and its real interpreter. You follow along and approve from your
own computer.

!!! note

    Can't install anything on the machine? Network gear, an ESXi host, a locked
    server: use a [target](../targets/overview.md) instead.

| | Device | Target |
|---|---|---|
| Ysra installed there | Yes | No |
| How I reach it | It connects out to Ysra | Your SSH or command-line tool, from your computer |
| Good for | Project folders, benchmarks, real hardware | Looking around, careful changes, runbooks |

## Add a device

<div class="steps" markdown>

### Install Ysra on it

```bash
curl -fsSL https://ysra.ai/install.sh | sh -s -- --device
```
Full steps: [Install on a server or edge device](../install/device.md).

### Ask to sign in

```bash
ysra login --device
```
The device shows a code such as `ABCD-EFGH`, valid for 10 minutes.

### Approve it from your computer

Desktop app → **Settings → Devices → Approve a device**. Type the code,
choose **Find it**, check the name and system, then **Approve**.

### Share a folder

```bash
ysra serve --folder /opt/app
```

</div>

![Approving a device](../assets/screens/desktop/device-approve.webp){ width="1242" height="962" }
/// caption
Approving a device by its code
///

> Diagram: A device signs in with a code you approve, connects out, and does the work on its own folder.

## Observe and change

Every device has a mode. You set it in **Settings → Devices**.

| Mode | What I do there |
|---|---|
| **Observe** (the default) | Look around and run checks. No edits |
| **Change** | Also edit files there, with your OK |

In observe mode, if you ask for a change I say: *"edge-01 is in observe mode;
switch it to change in Devices to let me edit there."* Switching is always your
action.

Devices often run something live. So heavy or disruptive work asks first in
either mode, and names the impact: a benchmark (*"this runs the model for about
2 minutes at full GPU"*), restarting the app, installing packages.

**Auto mode never covers changes on a device.**

## Work on a device folder from your laptop

1. On the device, keep `ysra serve --folder /opt/app` running (or install the
   [service](../install/device.md#5-keep-it-available-after-you-log-out-linux)).
2. In the desktop app, open **Settings → Devices**.
3. Choose **Folders on edge-01…**, then **New thread** next to the folder.

The thread behaves like any other: the working timeline, approval cards,
review, commit and push, undo. The work happens on the device.

![Folders shared by a device](../assets/screens/desktop/device-folders.webp){ width="1242" height="718" }
/// caption
A device's folders with New thread
///

You can also work directly on the device: SSH in, `cd` to the folder, run
`ysra`.

If the device loses its connection while I'm working, I pause and carry on when
it's back.

## Camera and GPU

By default my commands can't see any hardware on the device. The first time a
task needs it, I ask:

```text
Let my commands use the camera and GPU on edge-01?
```

![The hardware card in a device thread](../assets/screens/desktop/device-thread-hardware-card.webp){ width="1440" height="891" }
/// caption
Asking to use the camera and GPU on a device
///

- Asked once per conversation, for the hardware named.
- Lasts for that conversation only.
- Auto mode never answers it.

That's what makes real measurements possible: a before/after benchmark on the
actual GPU, or a detection run against the real camera.

## Remove a device

**Settings → Devices → Remove → Remove it.** The device is signed out at once.
Anything still running there stops being able to reach your work.

## Security: what a device can and can't reach

**A device can:**

- work on the folders you share from it with `ysra serve`, or the folder you
  run `ysra` in;
- run commands there in a contained space: writes only in that folder, no
  network except package registries during installs, no hardware without your
  OK.

**A device can't:**

- see or change your account, your balance, your sign-in or your other devices;
- reach your laptop's folders or your [targets](../targets/overview.md);
- be reached from outside. It only connects out; no port is opened on it.

**If a device is lost or compromised**, remove it in Settings. Its sign-in is
tied to that one device, expires on its own and is renewed regularly, and stops
working the moment you remove it.

The rest is the same as on your laptop: Git changes only when you ask, history
never deleted, undo is exact, and only what I read to do the work leaves the
machine. See [Security & privacy](../security.md).