Large language model-driven Multi-Agent Systems (MAS) face two key challenges in long-horizon tasks: collaboration workflows struggle to adapt to dynamic task states, and accumulated inter-agent interactions lead to lengthy, noisy contexts that degrade model performance and efficiency. To address these challenges, we propose MAS-OS, a collaborative Operating System (OS) for MAS. MAS-OS consists of three core modules: Memory Manager, Agent Scheduler, and Context Allocator. The Memory Manager maintains a structured memory pool and a global summary of agents’ interaction history. Based on the evolving memory states, the Agent Scheduler selects the next active agent, while the Context Allocator provides task-relevant memories to support effective context utilization. Experiments on four benchmark datasets demonstrate the effectiveness of MAS-OS. Compared with SOTA baselines, MAS-OS achieves an average improvement of 17.42% under the Llama-3.1-8B backbone. The code is available at https://anonymous.4open.science/r/MAS-OS-AD34.

Long video understanding requires reasoning over hours of content that far exceed the context window of multimodal large language models (MLLMs). Existing approaches compress video into fixed-length clips, yet events are fragmented and unrelated content is merged, yielding noisy retrieval. To overcome this challenge, we propose EMMA, an Event-driven Multimodal Memory system with Multi-Agent retrieval, which organizes memory around semantic events in three stages: (1) Event segmentation: a dual-stream detector adaptively locates event boundaries by fusing visual semantic drift with spoken transitions, rather than fixed windows. (2) Memory construction: built on the detected events, a three-level hierarchy of raw video, clip-level memory, and event-level memory stores content at progressively higher semantic abstraction. (3) Memory retrieval: a multi-agent framework collaboratively queries this hierarchy through query decomposition, tiered search, evidence validation, and answer generation, narrowing evidence from coarse events to precise clips. Extensive experiments on three challenging benchmarks demonstrate substantial improvements, showing the effectiveness of our event-aligned memory system. Code, prompt templates, and reproduction scripts are available at https://github.com/BAI-LAB/EMMA.

Large language models (LLMs) have made significant progress in various NLP tasks, especially with the rise of long-context models that handle extended inputs. However, the growing size of the key-value (KV) cache in Transformer architectures imposes heavy memory demands during decoding, becoming a major bottleneck. Existing sparse attention methods aim to address this issue but face two main limitations: (1) they introduce high overhead when retrieving important tokens for attention, and (2) they often fail to significantly reduce memory usage in practice. This paper presents Seer, a simple yet effective system for fast and accurate LLM decoding on long-context sequences. Seer leverages semantic space representations to manage the KV cache more efficiently. It dynamically estimates cluster-level importance to reduce both computation and memory access, while preserving output quality. Through offline training, Seer learns semantic centroids to enable fast token importance estimation. To further improve efficiency, it integrates product quantization with sparse attention for KV cache compression and uses an asymmetric distance lookup table to reduce computational cost. Evaluation on a diverse set of LLMs and tasks shows that Seer reduces theoretical attention bandwidth by 40× with minimal accuracy loss. Seer achieves a 5.57× attention speedup and a 2× End-To-End speedup over full attention at 32K sequence length. Our code is available at https://github.com/Acaciasama/Seer.