MunAI Offline — on your own computer
====================================

This folder runs MunAI on your computer instead of on someone's server. The
Harmonism articles are in here. The model that answers runs on your machine.
Your conversations stay in the browser on the device you use them from. After
the first setup nothing is sent anywhere.

There is also an installable app that does everything below without a window
of text: MunAI Offline for Mac, Windows and Linux, linked from
harmonism.io/munai-offline/ (Settings → Using MunAI offline). Use whichever
you prefer; this folder stays for anyone who would rather not install anything.


To start it
-----------

Mac:      double-click  MunAI Offline.command
Windows:  double-click  MunAI Offline.bat

The first run asks you to install one free program (Ollama, which runs the
model) and then downloads a model — between 6 and 27 gigabytes, once, depending
on how much memory your computer has. Later runs take seconds. A window of text
stays open while MunAI is running; that is normal. Closing it stops MunAI.

If macOS says the file is from an unidentified developer: right-click
MunAI Offline.command, choose Open, and confirm once.


To use it from your phone
-------------------------

Keep the window open on the computer. It prints two addresses; the second one
begins with a number, like http://192.168.1.14:8765/. Type that into your
phone's browser. The phone and the computer must be on the same Wi-Fi.

Two things to know about the phone. It works only while the computer is on and
this window is open — the model lives on the computer. And what you tell MunAI
on the phone is stored separately from what you tell it on the computer; they
are different browsers. Settings has Export and Import to move it across.

A phone does not run the model itself. The models small enough to fit on a
phone answer badly enough to misrepresent what they are speaking for, so that
option is not offered.


Which model it uses
-------------------

The launcher looks at your computer's memory and picks the strongest model
that fits with room for MunAI's whole doctrinal backbone:

  64 GB or more    Gemma 4 26B, uncensored, high precision   27 GB
  32 GB or more    Gemma 4 26B, uncensored                   17 GB
  16 GB or more    Gemma 3 12B, uncensored                    7 GB
  less             Qwen3.5 9B, uncensored                     6 GB

"Uncensored" means the refusals the manufacturer trained in have been removed,
so the model answers on contested ground — health, doctrine, history — instead
of declining. Everything else it learned is still there; MunAI's backbone
supplies the orientation.

The name of the model is written in model.txt in this folder. To use another,
put any name Ollama can pull in that file (gemma3:12b is a safe choice) and
start MunAI again. Larger models answer better and run slower.


What is in here
---------------

  MunAI Offline.command   the Mac launcher
  MunAI Offline.bat       the Windows launcher
  model.txt               which model to use (made on first run)
  app/                    the app itself, and the articles
  tools/serve.py          the small program that serves the app and passes
                          questions to the model; it logs nothing

Nothing in this folder phones home. There is no telemetry, no account, and no
update check unless you press the button for one in Settings.


If something goes wrong
-----------------------

The launcher says what is missing and what to do about it. The ordinary cases:

  Ollama is not installed yet — the launcher opens its download page.
  The model does not work with your Ollama — update Ollama from ollama.com;
    the launcher tells you when this is the problem.
  The computer went to sleep while the phone was connected — wake it and
    reload the page on the phone.

Everything else: harmonism.io/munai-offline/ has the same instructions in
the app itself, under Settings → Using MunAI offline.
