Memanto: Typed Semantic Memory with Information-Theoretic Retrieval for Long-Horizon Agents
Seyed Moein Abtahi, Rasa Rahnema, Hetkumar Patel, Neel Patel, Majid Fekri, Tara Khani
2026-04-27
Summary
This paper introduces Memanto, a new way to give AI agents a memory, and shows it works better than current methods.
What's the problem?
As AI agents become more complex and need to remember things over longer conversations, their 'memory' becomes a major problem. Current systems use complicated methods like knowledge graphs, which require a lot of computing power to store and retrieve information, slowing things down and adding complexity. They need to constantly analyze data, maintain the structure of the memory, and search through it using multiple steps.
What's the solution?
The researchers created Memanto, a simpler memory system for AI agents. Instead of a complex knowledge graph, Memanto uses a set of thirteen pre-defined categories to organize information. It also has a way to automatically resolve conflicting information and keeps track of changes over time. The key to its speed is a new type of database called Moorcheh's Information Theoretic Search, which finds information quickly – in under ninety milliseconds – without needing to 'index' the data beforehand, meaning there's no delay when adding new memories.
Why it matters?
Memanto is important because it significantly improves the accuracy of AI agent memory, achieving top results on standard tests. It does this while being faster, simpler to operate, and requiring less computing power than existing systems. This makes it more practical to build and deploy AI agents that can reliably remember and use information over extended periods, paving the way for more capable and useful AI.
Abstract
The transition from stateless language model inference to persistent, multi session autonomous agents has revealed memory to be a primary architectural bottleneck in the deployment of production grade agentic systems. Existing methodologies largely depend on hybrid semantic graph architectures, which impose substantial computational overhead during both ingestion and retrieval. These systems typically require large language model mediated entity extraction, explicit graph schema maintenance, and multi query retrieval pipelines. This paper introduces Memanto, a universal memory layer for agentic artificial intelligence that challenges the prevailing assumption that knowledge graph complexity is necessary to achieve high fidelity agent memory. Memanto integrates a typed semantic memory schema comprising thirteen predefined memory categories, an automated conflict resolution mechanism, and temporal versioning. These components are enabled by Moorcheh's Information Theoretic Search engine, a no indexing semantic database that provides deterministic retrieval within sub ninety millisecond latency while eliminating ingestion delay. Through systematic benchmarking on the LongMemEval and LoCoMo evaluation suites, Memanto achieves state of the art accuracy scores of 89.8 percent and 87.1 percent respectively. These results surpass all evaluated hybrid graph and vector based systems while requiring only a single retrieval query, incurring no ingestion cost, and maintaining substantially lower operational complexity. A five stage progressive ablation study is presented to quantify the contribution of each architectural component, followed by a discussion of the implications for scalable deployment of agentic memory systems.