AI Briefing#llm-memory
The latest AI and developer news about #llm-memory, with the original source and a short summary.
The latest AI and developer news about #llm-memory, with the original source and a short summary.
The system achieves 84.33% accuracy on FactKG using 108x fewer tokens than full-context inference while outperforming baselines on five of eight benchmarks.
The method injects numerical memory cartridges directly into Mistral-7B layers 7–31, bypassing text input and retrieval while maintaining bit-exact memory integrity.
The open-source tool achieves 89.8% Recall@1 on LongMemEval-S while running on minimal hardware like an Intel Celeron N4020.
Self-hosted AI assistant Frona has introduced a verified ontology-based memory system to reduce errors.
Performance and costs can be optimized by adjusting the amount of guidelines provided to agents based on model performance and scale.
Analyzed the KV cache capacity and memory requirements of the Qwen3.8-27B model based on context length.
Introducing mex, a tool that helps coding agents maintain accurate context by reflecting project changes.
It introduces GBrain, a knowledge layer that helps AI agents manage personal knowledge and generate citation-based answers.
HOLA, a model that introduces a hippocampal structure to prevent information loss occurring during the compression process of linear attention, has been released.
Microsoft has unveiled Memora, a memory framework that separates stored content from retrieval methods to enhance AI agents' ability to perform long-term tasks.
Stateful Swarms, which uses structured memory to lower AI agent costs and boost performance, has been released as open source.
AgingBench, a benchmark that measures the performance degradation of AI agents during long-term deployment, has been released.
A new open-source memory tool called ICM has been released to help maintain knowledge retention across sessions.
It is a local-first memory system that stores AI agent experiences as Git-version-controlled cards.
BrainAPI has been released, providing AI agents with structured memory through knowledge graphs that go beyond simple vector search.
The leak of Claude Code's source code has revealed the coding agent's memory architecture and subagent utilization techniques.