Desktop Services

Applications and setup guides for desktop-cachyos. Private links require Tailscale on your device. This hub is public; private services stay private.

Applications

Tailscale onlyOCRRead text from screenshots and documents. Choose GLM-OCR for images or Unlimited-OCR for images and PDFs. Models load on demand; unload them to free GPU memory. Public page · authenticated uploadsFile uploaderSend files to the desktop using your existing upload token. Files are saved in ReceivedFiles. The token is never included in this hub. Tailscale onlyMinecraft managerSign in to MCSManager to manage instances, read consoles, and inspect resources. The panel and daemon share port 10000.

Opening private pages

Connect to Tailscale first. If Firefox reports a secure-connection error for this desktop, add desktop-cachyos.tailb60e44.ts.net to its DNS over HTTPS exceptions so Tailscale can resolve the private address. HTTPS remains enabled.

OCR is served at its own private HTTPS address on port 8090. The previous public /ocr/ route has been retired.

Kate AI Coder

Tailscale API · on demand

Edit a selected section of code in Kate with the desktop's Qwen3-Coder model. The desktop runs the model; the laptop only needs Kate and the AI Edit Selection plugin. This is an editor integration, not a website.

1. Start the model on the desktop

Run these in a desktop terminal. The service stays off after login until you start it; the first start may download the large model and take time.

~/.local/bin/kate-ai-coder on
~/.local/bin/kate-ai-coder endpoint
~/.local/bin/kate-ai-coder logs

The installed launcher binds to the desktop's Tailscale address when available. Use http://100.72.10.76:8080/v1 in both desktop Kate and laptop Kate. HTTP here travels inside the encrypted Tailscale connection. A started service can still be loading: check readiness from a Tailscale-connected device:

curl --noproxy '*' http://100.72.10.76:8080/health

2. Set up Kate on the laptop

Connect the laptop to Tailscale. Copy the existing plugin package from the desktop over SSH, extract it, and run its installer. SSH may ask for your desktop credentials; the installer lists any missing build dependencies.

mkdir -p ~/Downloads
scp bradleyy@100.72.10.76:/home/bradleyy/Downloads/kate-ai-edit-v0.1.5.zip ~/Downloads/
unzip ~/Downloads/kate-ai-edit-v0.1.5.zip -d ~/Downloads
cd ~/Downloads/kate-ai-edit
bash install.sh
~/.local/bin/kate-ai-edit-configure http://100.72.10.76:8080/v1

The desktop model service and helper commands are already installed. A desktop plugin file is also present; if AI Edit Selection is missing from Kate's plugin list, run the same package installer on the desktop. Configure desktop Kate with that same full URL:

~/.local/bin/kate-ai-edit-configure http://100.72.10.76:8080/v1

3. Edit a selection

  1. Restart Kate after installing the plugin. Open Settings → Configure Kate → Plugins and enable AI Edit Selection.
  2. Select a normal section of code, press Ctrl + Alt + I, and describe the change.
  3. The plugin replaces that selected range. Review the result; use Kate's undo if needed. It sees the selection and limited file context, not your whole project.

When finished, stop the model on the desktop to release RAM/VRAM. Use the OCR page's unload control before starting a large coding session if you need more GPU memory.

~/.local/bin/kate-ai-coder off

Moonlight & Sunshine

Use the Tailscale addressSunshine settingsPair Moonlight and manage streaming applications. Sunshine has its own HTTPS certificate and login. Its listeners are not restricted to Tailscale interfaces.
Desktop helpersHeadless displayA virtual display and two watchdogs keep remote streaming usable. These are background helpers, not extra web applications.

On the laptop, connect to Tailscale, open Moonlight, and add 100.72.10.76 as a PC. Enter Moonlight's pairing PIN in Sunshine's settings. Use the desktop's existing headless controls when needed:

~/.local/bin/sunshine-headlessctl status
~/.local/bin/sunshine-headlessctl headless
~/.local/bin/sunshine-headlessctl local

Local developer tools

Desktop localhost onlyOpenCode embeddingsQwen3 embedding API for local research memory at http://127.0.0.1:8081/v1. There is no public or Tailscale proxy for this API.
Desktop utilityPenguinBurnerHardware controls backed by a local privileged daemon. Use the desktop application; it has no web endpoint.
API and service notes

The file uploader accepts authenticated POST /upload requests. Existing root-host upload links using a #send=… fragment still open the uploader.

The MCSManager daemon uses wss://desktop-cachyos.tailb60e44.ts.net:10000 with Socket.IO path /mcsm-daemon/socket.io. Open the panel for normal use.

The OCR model API stays on desktop localhost port 18080; Unlimited-OCR uses 18082. The OCR page controls both. Steam's localhost port 8080 is separate from Kate's Tailscale port 8080.

These instructions describe installed services and their controls; this static page does not start models or report live status.

Private application links require Tailscale. Keep upload tokens and application credentials private. Model and streaming controls run on the desktop unless a step explicitly says laptop.