Reproduce “Agents in the Large: Perception-Centered Architecture for Persistent Agents” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Shihan Dou, Haoxiang Jia, Shichun Liu, Feng Chen, Chenhao Huang, Yujiong Shen, Shaofan Liu, Jiayi Chen and 9 more

Pera introduces a perception-centered architectural framework for AI agents to provide continuous assistance in dynamic, long-lived settings. This framework extends cognitive language agents with lifecycle perception and a control plane, enabling proactive identification and execution of self-maintenance tasks in response to changing internal and external conditions.

Key insights

  • Persistent agents require active perception to continuously recognize service-relevant signals from both internal states and external environments, beyond the scope of a single active task objective.
  • A dedicated Control Plane is crucial for translating these perceived signals into actionable 'Lifecycle Task Packages' that drive agent self-modification and adaptation.
  • The ability to distinguish and manage 'Lifecycle Tasks' separate from 'Episodic Tasks' is fundamental for agents to proactively maintain and improve their capabilities in dynamic settings.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.