Qontext: A Quipu-Inspired Conversational Memory for Local Language Models

Abstract

A conversational memory for language models that stores short, self-contained third-person facts ("knots") and returns only those a given prompt needs, inspired by the Inca quipu. On 800-turn conversations with human-written dialogue as filler, a 12B model answers 9.7/10 planted-fact questions from a 116-token pack against 9.3/10 from a 13,853-token transcript: tied accuracy at a 119-fold reduction in prompt tokens. On smaller models the reduction also improves accuracy, but the effect vanishes at 12B, so it is stated conditionally. The paper additionally reports a decomposition of retrieval failure showing that 79.8% of missed facts on conversational-turn queries are unreachable rather than mis-ranked, and five negative results against that population, including contextual sentence embeddings. Code, benchmarks and all measurements are released, including four occasions on which the authors' own instrumentation produced false findings.