A developer benchmarked a homelab agent memory stack on the LoCoMo benchmark and improved accuracy from 63% to 80% by adding hybrid search with BM25 and a local reranker. The developer found that dens…
Recordari, a custom memory graph built by developer Corby, scored 0.831 on the LoCoMo benchmark, outperforming Microsoft Research's Memora, which published a score of 0.801. The benchmark, run with th…
The BEAM benchmark, developed by the team behind Suprflo, evaluates AI agent memory over long, multi-session histories with changing facts, ranging from 100K to 10 million tokens. It measures distinct…
Backboard, a Canadian AI infrastructure company, reported a 99.95% score on the LoCoMo memory benchmark by post-training memory directly into model weights, rather than using retrieval. The company pr…
A July 2026 paper (arXiv 2607.01935) identifies 'ghost memory' as a state coordination failure in agent memory systems, where old, current, and transition facts coexist and mislead the model, and repo…
Researchers introduced AgentMemBench, a benchmark evaluating five long-term memory strategies for conversational AI agents across three datasets and 491 annotated question turns, finding that external…
Researchers introduced CoMem (Comprehension Memory), a method that leverages the depth division of labor in large language models to enable unbounded-context memory. In evaluations, CoMem achieved 97.…
Researchers propose SF-AMS, a framework for strategic forgetting in LLM agent memory systems that replaces static retrieval with a utility-driven survival mechanism, achieving consistent gains over ba…
Researchers introduce MemHop, a multi-hop memory benchmark of 1,000 questions at hop depths 1-5 across 10 social-network scenarios, and ProGraph, a two-layer memory architecture combining profile expa…
A new training-free framework, Memory--Skill Co-Evolution (MSCE), enables long-horizon LLM agents to convert passive memory traces into reusable skills, outperforming state-of-the-art baselines on Evo…
Researchers introduce MOSAIC (Memory-Organized Structured Agent for Information Collection), a structured, conflict-aware long-term memory framework for LLM agents that achieves 89.35% accuracy on the…
Researchers introduced PPRO (Profile-guided Personalized Retrieval Optimization), a framework that enhances memory retrieval for long-term conversational agents by integrating user profiles. The syste…
Researchers introduced NapMem, a framework that transforms long-term user memory into a structured action space for conversational agents, enabling active navigation across memory granularities. Train…
Microsoft researchers introduced Memora, a scalable memory system for AI agents that decouples stored content from retrieval mechanisms, achieving state-of-the-art performance on long-horizon benchmar…
Mem0, an open-source memory layer for AI agents with nearly 60,000 GitHub stars, has released v3.0 with temporal reasoning and multi-tenant isolation. The project scored 94.8 on LongMemEval and 91.6 o…
Signet launched a new open-source tool that preserves agent identity, memory, and secrets across model switches, storing state outside any single model or harness. The tool automatically distills sess…
Researchers introduced Eywa, a provenance-grounded long-term memory architecture for AI agents that stores immutable source evidence before deriving facts, achieving 90.19% judge accuracy on the LoCoM…
Writer's research paper reports that memory systems can amplify sycophancy by up to 25x, but the amplification is traced to two design decisions: an evaluation prompt ordering the model to answer sole…
Memory systems in AI assistants are evolving beyond the short-term versus long-term split, with OpenAI and LangGraph SDKs advocating for a three-tier stack of working memory, durable state, and retrie…
Backboard has launched Backboard Development Studio, an open beta coding harness that uses a memory-first architecture to outperform frontier AI models while relying on open-source models. The system,…