RhyGPU · local-first AI

I build software I can actually see inside.

Local-first AI tools and memory systems — inspectable, stateful, and mine. From the engine room, the writing desk, and the terminal.

3
active projects
11
devlog entries
Rust
+ TS, C++, Electron
Local
first, always
ACTIVE ALPHA

Mnemosyne

A desktop AI narrative engine that separates character memory, world state, prompt context, and debugging traces so LLM-driven characters can change without losing continuity.

TauriReactTypeScriptRust
Mnemosyne home dashboard showing living worlds, recent sessions, and engine status.

OmniPlanner

ACTIVE APP

An Electron desktop app for personal planning workflows, built around local data, weekly structure, goals, habits, and optional AI assistance.

ElectronViteReactTypeScript
GitHub →

Pythagorean Harmony

PLAYABLE PROTOTYPE

A UE5 prototype exploring math-based combat, where enemies are defeated by transforming equations with operation weapons and executing with the correct value of x.

Unreal Engine 5C++Blueprint tuningThird Person Template
GitHub →

Latest devlogs

Open full journal →
010Prompt Cleanup V2How Mnemosyne moved from adding more prompt rules to cleaning the narrator brief, preserving character agency, and separating player paper UI from terminal dev machinery.009Starting Fresh Without Erasing the SoulFresh scenario state, stale world cleanup, and the first reset boundary.008Making the Session InspectableVisible chat export, payload history, and the first real state diagnosis.007Making Bad Outputs RecoverableUX comfort, regenerate/fix response, and the road to real branching.006Opening the Black BoxLLM Payload Inspector, Latest Exchange, and the tail excerpt bug.