Overview

Welcome back! Last time in January we built companions with RPMax. A lot has changed since then. The local LLM roleplay scene has evolved fast. New models, better techniques, stronger community. Tonight we're using Magidonia-24B-v4.3 by TheDrummer - a 24B model used for roleplay in this session.

We're skipping the setup. You already have Ollama and Open WebUI from previous meetups. Tonight we dive straight into the good stuff. No waiting around for installations. We have less than two hours, and we've got character creation, working examples, and parameter tuning to cover. Let's

make it count.

What You'll Learn

This tutorial takes you from "model downloaded" to "shipping a chatbot that actually works." Here's the map:

The model itself: why Magidonia beats everything else in its weight class
Configuration: the parameters that matter, and the ones that don't
Character creation: how to write a good system prompt that Magidonia respects
Working examples: copy-paste prompts you can use tonight and take home
Community patterns: techniques from the 9,000-person BeaverAI Discord

You'll understand not just what to do, but why each step exists. We're going deep but keeping it conversational - like a smart friend explaining things over coffee.

What You'll Need

Hardware:

Ollama installed (you have this)
Open WebUI running (you have this)
16GB+ RAM minimum (32GB+ recommended for 24B models without compromise)
About 25GB free disk space

Time:

90 minutes if you follow along
30 minutes if you want to skip to the examples

Software:

- Same stack as January: Ollama + OpenWebUI

No new dependencies
Everything runs local

The Model Swap: RPMax → Magidonia

If you attended January's RPMax tutorial, some stuff will look familiar. Architecture decisions? Different. RPMax uses ArliAI's base and chat template. Magidonia uses Mistral's architecture with TheDrummer's fine-tuning. RPMax was bulletproof for instruction-following. Magidonia is specialized for creative work - roleplay, storytelling, character consistency.

The community is different too. RPMax lives in ArliAI spaces. Magidonia lives in BeaverAI - 9,000+ Discord members, constantly merging and iterating. The conversation is real-time.

One big practical difference: we don't need a custom Modelfile for basic chat. The model comes with the Mistral V7-Tekken chat template baked in. Ollama handles it automatically. We used to need to write YAML for this stuff. Now we don't.

Our Philosophy

Everything local. No cloud. No subscriptions. No one monitoring your conversations. You own your data. You own your chatbots. If something breaks, you fix it. If you want to modify the model, you can.

That's the deal. It means more setup than ChatGPT. It means you're responsible for your hardware. It's worth it.

Let's Go

By the end of this tutorial, you'll have a character prompt, local model settings and a way to test whether the character stays consistent as the chat grows.

Let's build something real.