DMF: A Deterministic Memory Framework for Conversational AI Agents
The Deterministic Memory Framework (DMF) aims to enhance memory systems for conversational AI agents. It replaces traditional LLM-based summarization with a deterministic approach that reduces token costs and improves coherence. Experimental results demonstrate DMF's efficiency, achieving comparable accuracy while significantly lowering memory management costs.
- ▪DMF utilizes a CPU-first approach grounded in classical NLP analysis and vector geometry.
- ▪The framework assigns a Survival Score to each interaction, governing relevance as new interactions occur.
- ▪DMF achieves up to 242 times fewer tokens used over entire conversations compared to existing memory layers.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2606.03463 (cs) [Submitted on 2 Jun 2026] Title:DMF: A Deterministic Memory Framework for Conversational AI Agents Authors:Matteo Stabile, Enrico Zimuel View a PDF of the paper titled DMF: A Deterministic Memory Framework for Conversational AI Agents, by Matteo Stabile and 1 other authors View PDF HTML (experimental) Abstract:Conversational AI agents require memory systems that are both scalable and semantically coherent across long interaction horizons. Existing approaches rely predominantly on large language model (LLM)-based summarisation at write time, which introduces non-determinism, escalating token costs, and opacity in pruning decisions.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
