Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The paper introduces Autogenesis Protocol (AGP), a novel framework for self-evolving LLM-based agents that decouples resource management from evolution mechanisms. AGP's Resource Substrate Protocol Layer (RSPL) explicitly models agents, tools, prompts, environments, and memory as versioned resources with defined lifecycles, while the Self Evolution Protocol Layer (SEPL) provides a closed-loop interface for proposing, assessing, and committing improvements. The authors instantiate AGP in the Autogenesis System (AGS) and demonstrate its effectiveness on complex benchmarks, showing consistent improvements over existing agent systems via dynamic resource refinement.
Monolithic LLM agent systems are out: Autogenesis offers a self-evolving protocol that dynamically refines agents, tools, and prompts for consistent performance gains.
Recent advances in LLM based agent systems have shown promise in tackling complex, long horizon tasks. However, existing agent protocols (e.g., A2A and MCP) under specify cross entity lifecycle and context management, version tracking, and evolution safe update interfaces, which encourages monolithic compositions and brittle glue code. We introduce \textbf{\textsc{Autogenesis Protocol (AGP)}}, a self evolution protocol that decouples what evolves from how evolution occurs. Its Resource Substrate Protocol Layer (RSPL) models prompts, agents, tools, environments, and memory as protocol registered resources\footnote{Unless otherwise specified, resources refer to instances of the five RSPL entity types: \emph{prompt}, \emph{agent}, \emph{tool}, \emph{environment}, \emph{memory} with agent \emph{outputs}.} with explicit state, lifecycle, and versioned interfaces. Its Self Evolution Protocol Layer (SEPL) specifies a closed loop operator interface for proposing, assessing, and committing improvements with auditable lineage and rollback. Building on \textbf{\textsc{AGP}}, we present \textbf{\textsc{Autogenesis System (AGS)}}, a self-evolving multi-agent system that dynamically instantiates, retrieves, and refines protocol-registered resources during execution. We evaluate \textbf{\textsc{AGS}} on multiple challenging benchmarks that require long horizon planning and tool use across heterogeneous resources. The results demonstrate consistent improvements over strong baselines, supporting the effectiveness of agent resource management and closed loop self evolution.