Home

Welcome to the Storyteller Wiki

Note This project was set-up to help me learn more about LLM application design, prompting, memory shaping, and validation behavior, not to spend most of the project inside framework infrastructure. In the end it became part of a short blog-post I've written.

Storyteller is a framework-light Java CLI storytelling application that communicates with OpenAI-compatible chat endpoints (running locally or remotely via backends like LM Studio, Jan.ai, llama-server, or mlx-vlm).

Rather than relying on heavy frameworks like Spring Boot or Quarkus, it is deliberately designed to keep runtime and source layouts small, transparent, and easy to inspect.

Core Features

• Framework-Light Design: Built with pure Java 25+ and Maven for fast execution and minimal runtime footprint.
• OpenAI-Compatible Integration: Connects seamlessly with local LLM servers or hosted APIs.
• Multi-Layer Derived Memory: Manages short-term history alongside background summaries and a canonical state file.
• Sequential Background Tasks: Runs memory updates asynchronously through a dedicated single-threaded queue.
• CLI Customizations: Powered by JLine with built-in shortcuts for story undo, continuation, and rules reset.

Quick Links

• Installation & Running: Prerequisites, Maven packaging, native builds, and run commands.
• Configuration & Hardware Guide: Hardware fit tables, memory usage, local model setup, and .config parameters.
• Architecture & Shortcuts: Terminal key bindings, command reference, prompt assembly, and package layout.
• Limitations / bottlenecks: Entity Contagion & Drift and other limitations like Lexical Corruption & Quantization Noise
• Limitation mitigations: How Storyteller Mitigates Local LLM Limitations
• Changelog

---

Repository: jbrugman/Assistant