AetherCompute — lend your graphics card to AI
=============================================

Your computer's graphics card runs AI language models (the kind behind chat
assistants) for other people while it would otherwise sit idle.

For now this is a test: your computer connects, sets itself up and shows up
on the organiser's screen. No real requests are sent to it yet, and nothing
is paid.

If you downloaded this from the AetherCompute website, there is nothing to
type in: it finds our server by itself, through the website. Otherwise you
are asked for the address you were given, once, the first time you start.


WHAT YOU NEED
-------------
  * A graphics card with 8 GB of its own memory, or a Mac with an Apple chip
    (M1 or newer) and 16 GB of memory. Other computers can join the test too;
    they just won't be given models to run.
  * About 2 GB of free disk.

You do NOT need to install anything, not even Python. The first time you
start, it sets itself up:

  * its own copy of Python, 25-75 MB, kept inside this folder
  * a few small packages the node needs, also inside this folder
  * llama.cpp, the program that runs the models, in the build that suits your
    graphics: about 12 MB on a Mac, about 30 MB for AMD or Intel graphics,
    about 580 MB for an NVIDIA card. It goes in a hidden ".aether" folder in
    your home folder, and is set up in the background while the node runs.

Every download is checked against a fingerprint before it is used. Nothing
is installed system-wide, nothing appears in your applications, and any
Python you already have is left alone.


STARTING IT
-----------
Unzip the download first. On Windows, right-click it and choose "Extract
All"; starting it from inside the zip won't work.

WINDOWS
  1. Double-click  start.bat
  2. If a blue "Windows protected your PC" box appears, click "More info",
     then "Run anyway". If Windows asks whether to run the file, click "Run".

macOS
  1. Double-click  start.command
  2. The first time, macOS stops it, because it was downloaded. Open System
     Settings, then Privacy & Security, scroll down, click "Open Anyway" next
     to start.command, then "Open". After that, double-clicking just works.

LINUX
  1. ./start.sh
     (If it says permission denied:  chmod +x start.sh)


WHAT YOU SHOULD SEE
-------------------
The first time only: "Setting up Python (once…)" with a download bar, then
"Installing what the node needs". That takes a minute or two. If the
connection drops, run it again and it carries on where it stopped.

Then a panel showing your computer, headed "CONNECTED — LANGUAGE MODELS",
with a line for language models. While llama.cpp is being set up:

    Language models       setting up llama.cpp 42%

and once it is ready:

    Language models       ready · NVIDIA GeForce RTX 4070 12 GB · 0 on disk

If llama.cpp can't run on your computer at all, that line says why. Whether
your graphics are good enough to be given models is decided on the
organiser's side. Either way the node keeps running.


WHEN OUR SERVER MOVES OR IS SWITCHED OFF
----------------------------------------
Downloaded from the website: nothing to do. Our server's address changes
from time to time, and the app finds the new one by itself, through the
website. While our server is switched off, the app says so and waits, then
connects by itself once it's back.

Given an address to type in instead: when it changes, start once with the
new one:

    Windows   start.bat  https://the-new-address
    macOS     ./start.command  https://the-new-address
    Linux     ./start.sh  https://the-new-address

It is remembered after that.


STOPPING
--------
Ctrl+C in the window, or close it.


REMOVING IT
-----------
Delete this folder, which takes its Python and packages with it. Then delete
the hidden ".aether" folder in your home folder, which holds llama.cpp and
any models the node downloaded.
