LLMs are General Asynchronous Agents

2026-09-28

AI Summary

This study addresses the real-time concurrency bottleneck faced by large language models (LLMs) in speech and embodied scenarios due to their inherently sequential interaction mechanisms. To overcome this limitation, we propose a universal asynchronous agent framework that leverages inference coroutines and overlapping memory state management to enable concurrent multi-task adaptation. This architecture allows Qwen3.x models to natively support asynchronous tasks without requiring task-specific training. We validate the framework’s general-purpose asynchronous processing capabilities across diverse applications, including streaming video understanding, video games, and surveillance systems. Ultimately, this work establishes an efficient paradigm for deploying LLMs in complex, real-time interactive environments.

Abstract

Modern LLMs are increasingly capable as autonomous agents, but they follow sequential interaction cycles: read, think, reply or call tools, repeat. Many real-world use cases are not sequential: voice assistants, embodied agents, and monitoring systems receive new inputs while they think or perform another task. Modern LLMs address this with specialized architectures for voice interaction and video streams, VLAs for robot control, asynchronous tool calling for API usage, and others. In this work, we generalize from different asynchronous tasks to general asynchronous agents that can adapt to different types of concurrency. To achieve this, we develop an asynchronous LLM framework that lets users (or the agents themselves) define inference coroutines with overlapping memory states. We showcase that Qwen 3.x models are capable of asynchronous operation for streaming video understanding, videogames, and monitoring, without task-specific training.

Problem

• Large Language Models
• Asynchronous Agents
• Concurrency
• Autonomous Agents
• Sequential Interaction

Innovation

• Asynchronous Agents
• Inference Coroutines
• Overlapping Memory States
• General Asynchronous Framework
• Large Language Models