Basics

Guides

API Reference

Menu

Basics

Guides

API Reference

Hogan: the AI Assistant in AGR

Hogan is an assistant built into AGR. You type a request in plain English and Hogan writes Aussom code for it, or points you at the right documentation page. It is there to shorten the gap between "I want a form with a dropdown" and having one on screen.

This guide explains what Hogan can do, how to install it, and the words that tell it what you want.

What Hogan Is

Hogan is a small AI model trained specifically on Aussom. Three things about it are worth knowing up front:

It runs on your computer. Once you have downloaded it, Hogan needs no internet connection and sends nothing anywhere. Your code and your questions stay on your machine.

It is small and specialised. Hogan is about a thousand times smaller than the large assistants you may have used on the web. That is deliberate - it fits on a laptop and answers in seconds. The trade is that it knows Aussom and little else. Ask it to write Aussom and it does well; ask it about world history and it will not.

It can be wrong. Hogan writes code that usually works and sometimes does not. AGR checks the code before running anything for you, and tells you what it found. Read what it writes, the same as you would with code from a web search.

Installing the Model

Hogan does not ship with AGR, because the model file is large. You download it once.

  1. Open Help > Hogan Configuration.
  2. You will see a list with one entry, Hogan 0.7 (Q4_K_M).
  3. Click Download.

The file is about 940 MB, so it takes a few minutes on a normal connection. A progress bar shows how far along it is, and you can press Cancel at any time - a cancelled download leaves nothing behind and starts over cleanly next time.

When it finishes, AGR checks the file against a known fingerprint. If anything went wrong in transit it says so and discards the file rather than handing you a half-downloaded model that misbehaves in confusing ways.

Hogan switches on as soon as the download finishes - there is nothing else to set.

The model is saved in ~/.agr/models and is shared by every AGR window, so you only ever download it once no matter how many projects you work on.

Already have a model file? Click Browse... in the same dialog and point AGR at any .gguf file on your disk.

That is the whole setup. Downloading a model turns Hogan on, so there is nothing to configure and no file to edit. Open the Hogan Chat tab and ask it something.

Warming Up

Open the Hogan Chat tab and Hogan starts loading straight away, before you have typed anything:

Hogan is warming up ...
Hogan is ready (2.1s). I am your Aussom assistant: I can write Aussom code
for the REPL or the code window, and find the documentation for a class
when you ask for it.

This happens once per AGR session. It is done while you are reading the pane, so your first question does not have to wait for it. Loading is usually a second or two, though the very first time after downloading can take longer while your computer reads the file from disk for the first time.

After that the model stays in memory for as long as AGR is running, and answers come back in seconds.

While Hogan is answering you will see:

Hogan:
Working on it ...

Then the answer appears a piece at a time as it is written, so you can start reading before it finishes.

Asking for Code

Just describe what you want:

Make me a login form.

Hogan writes the program and shows it in the chat. Nothing runs by default - you see the code first and decide what to do with it.

Sending the code somewhere

The words you use decide where the code goes. There are three places.

What you say Where the code goes
nothing special printed in the chat, for you to read
"...and run it" runs straight away in AGR
"...in the code window" loaded into the Code tab, ready for you to press Run

So:

Make me a login form.                          - shows it
Make me a login form and run it.               - runs it
Put a red button in the code window.           - loads it into the Code tab

The exact phrases that trigger each one are:

  • Run it: "run it", "run this", "run that", "execute it", "execute this", "execute that"
  • Code window: "code window", "code box", "in the editor", "to the editor", "into the editor", "load it", "load this"

Anything else prints in the chat. This is deliberately strict. Asking to see code and having it run instead would be a nasty surprise, so AGR only runs things when you have clearly said to.

Describing what code does is not the same as asking to run it. "Write a program that runs a report" prints in the chat, because "runs a report" is a description, not an instruction.

Copying code out

Right-click any code Hogan has written in the chat. A small menu offers:

  • Copy - puts the whole program on your clipboard
  • Set in Code Window - drops it into the Code tab

What AGR checks before running anything

When you ask Hogan to run code or load it into the code window, AGR tries it first in a safe, separate space. If it works, you get it. If it does not, you still get the code, along with a plain explanation:

  • "This does not compile, so it will not run anywhere." The code has a mistake in it. Ask Hogan again, or fix it yourself.
  • "This was blocked because it uses an API that is not allowed there." The program tried to reach outside the user interface - files or the network, for instance. It was not run for you.
  • "This could not be run automatically, but it may still run in the current Engine without issue." The safe space does not have the variables you have already created, so code that uses them fails there and may be perfectly fine here. Press Run and see.

The code is always shown, whatever the check found. Being told what happened is more useful than being handed nothing.

Asking About the Documentation

Ask a documentation question and Hogan hands you links instead of prose:

Can you get me the doc on TextField.

You get a short list of matching pages. Click any name to open it in AGR's documentation browser.

This is on purpose. A documentation page is thousands of words, and pasting one into a chat window buries the conversation. A link costs one line and opens in the window built to display it.

Phrasing that works well:

Can you get me the doc on ComboBox.
Show me the docs for Button.
What are the docs on VBox.

Searching finds JavaFX classes first, since that is what AGR is for. If you want a different toolkit, name it - "gtk dropdown" returns gtk pages.

Just Talking

Hogan handles ordinary conversation too:

You: Hi Hogan, I'm Alex, nice to meet you.
Hogan: Hi Alex, how can I help you?

You: What can you do?
Hogan: I answer questions about Aussom and write Aussom code.

Nothing runs and no code appears. Hogan only writes code when you ask for code.

Hogan Across Several Windows

AGR works on one folder at a time, and you can have several windows open on different projects. The model is shared by all of them - it is downloaded once and lives in ~/.agr.

One thing to know: if you change your model in one window, the window you changed it in picks up the new one right away, but any other window that is already open keeps the model it loaded until you restart it.

If Something Goes Wrong

"No model is configured." You have not downloaded a model yet. Open Help > Hogan Configuration and click Download.

The download failed. The message says why. "Not in the repository" means the model list is out of date - update AGR. Anything about the network means try again; a failed download never leaves a broken file behind.

Hogan writes code that does not work. It happens. Ask again with more detail - naming the controls you want helps - or fix the code yourself in the Code tab. Hogan is a starting point, not a finished product.

Hogan says something odd about itself. It is a very small model, and questions outside Aussom are not what it was built for. Keep to code and documentation and it is on much firmer ground.

Where to Go Next

  • AGR Overview - the rest of the window and how workspaces work.
  • AGR Coding Guide - writing Aussom that draws to the graphics pane.