LoLCoach for Windows
A League of Legends coach that runs entirely on your PC. This repository hosts the prebuilt Windows application.
Source code, screenshots and documentation: https://github.com/Mattbusel/lolcoach
What this is
LoLCoach downloads your own ranked games from Riot, reconstructs what happened in them (wave states, jungle paths, why each death happened, where the game swung), and lets you talk to a locally fine-tuned model about it. Nothing is sent anywhere. The model runs on your machine, so conversation is unlimited and costs nothing.
Download and run
- Download every file in this repository (see below).
- Keep the folder together. Double click
LoLCoach.exe. - Paste a free Riot API key from https://developer.riotgames.com
- Add your Riot ID as
GameName#TAG, then press Sync.
The fastest way to get all of it:
pip install huggingface_hub
huggingface-cli download Fungle/lolcoach-windows --local-dir LoLCoach
Then run LoLCoach\LoLCoach.exe.
Size
About 22.6 GB. Most of that is the Qwen2.5-7B base model, which ships inside the bundle so the app works offline once downloaded.
Requirements
To review your games, no GPU needed: Windows 10 or 11 64 bit, 8 GB RAM, 8 GB free disk, and a free Riot API key.
To chat with the AI coach: NVIDIA GPU with 8 GB VRAM or more (RTX 3060 or 4060 and up), 16 GB RAM, 20 GB free disk, recent driver with CUDA 12.
Without a supported GPU the full match review still works. Only the chat is slow or unavailable. The app checks your own machine: open it and click Requirements.
What is in the bundle
| Part | Contents |
|---|---|
LoLCoach.exe, _internal/ |
The app and its Python runtime |
models/ |
Qwen2.5-7B-Instruct base model and the sentence-transformers embedding model |
data/models/ |
The LoRA adapter fine tuned for League coaching |
data/index/ |
FAISS retrieval index over Riot static data, patch notes and the League wiki |
data/assets/ |
Summoner's Rift minimap and champion portraits from Data Dragon |
The bundle deliberately ships with an empty match library. You sync your own games on first run.
Licensing
MIT covers the application code. The base model is Qwen2.5-7B-Instruct under its own licence. Retrieval content is Riot static data (Riot terms) and the League of Legends Wiki (CC BY-SA 4.0, attribution preserved in the app).
League of Legends is a trademark of Riot Games, Inc. This project is not affiliated with or endorsed by Riot Games.