Cheeney's GoLLM

Abstract

Cheeney’s GoLLM (Goal-Oriented Large Language Model) was an exploratory multi-pass LLM orchestration architecture created by Christopher Cheeney in March 2024 using the FluxPrompt visual development environment. The system was designed to simulate persistent purpose modeling and structured internal dialogue. Its ten-agent-step, twenty-node architecture identified the subject of a conversation, inferred why the user asked the question, generated four possible motivations for responding, selected and refined a guiding objective, produced a candidate response, and evaluated that response through model-assessed truthfulness, necessity, and kindness gates. Failed responses triggered regeneration from the original prompt. GoLLM was technically an orchestration module wrapped around an existing LLM, not a newly trained foundation model. In the author’s qualitative assessment, the additional inference passes produced somewhat more coherent, purposeful, and context-sensitive responses. The observed improvement was marginal, however, relative to the added inference calls, latency, platform credits, state management, and architectural complexity. The development branch was therefore abandoned. This technical note preserves the architecture, implementation history, observed negative engineering result, limitations, and surviving evidence. The original FluxPrompt project remains inspectable and executable. It was successfully executed again on August 23, 2026, using GPT-4o; that later model assignment should not be interpreted as identifying the model used during the original March 2024 experiment. This publication is intended to establish a clear historical and technical record of the implementation and its outcome, not to claim that GoLLM constituted a major or novel advance over the broader field.