https://github.com/anatta336/llm-dungeon

Using LLMs for a text game
https://github.com/anatta336/llm-dungeon

Last synced: 6 months ago
JSON representation

Using LLMs for a text game

• Host: GitHub
• URL: https://github.com/anatta336/llm-dungeon
• Owner: Anatta336
• Created: 2024-09-08T20:42:11.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2024-10-13T16:49:21.000Z (12 months ago)
• Last Synced: 2025-02-13T22:44:24.532Z (8 months ago)
• Language: Go
• Size: 4.55 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: readme.md

Awesome Lists containing this project

README

          
# Dungeon

An LLM-powered text adventure game that attempts to have a coherent world model.

![image](https://github.com/user-attachments/assets/3c89a7f3-89f0-4496-9f31-96fa93780302)

## Status

Imperfect proof of concept.

## Quickstart

- Have Ollama installed and set up.

- `ollama serve`

- `. ./build.sh && go build -o bin/main main.go && bin/main`

## Process structure

```mermaid

graph TB

user(["User"])

llmAdjudicate[["LLM: Adjudicate"]]

world[("World model")]

llmEncode[["LLM: Update State"]]

user --"Freeform input"--> llmAdjudicate

world --> llmAdjudicate

llmAdjudicate --"Freeform description"--> llmEncode

world --> llmEncode

llmEncode --"Update"--> world

llmAdjudicate --"Freeform description"--> user

```

## Plan

- Do not rely on LLM to update JSON directly.

    - Instead describe changes it wants to make and apply those.

- Adjudicator to stop inventing objects.

- Adjudicator to be less "helpful" so it doesn't just solve problems.

- Generation of scenes.

    - Goal.

    - Obstacles.

    - Means of overcoming obstacles.

- Some kind of player health/condition tracking?

    - Something more interesting that can take advantage of LLM reasoning.

- Anything to optimise LLM steps.