Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OpenSage: Self-programming Agent Generation Engine

Hongwei Li, Zhun Wang, Qinrun Dai, Yuzhou Nie, Jinjun Peng, Ruitong Liu, Jingyang Zhang, Kaijie Zhu, Jingxuan He, Lun Wang, Yangruibo Ding, Yueqi Chen, Wenbo Guo, Dawn Song

TL;DR

OpenSage is proposed, the first ADK that enables LLMs to automatically create agents with self-generated topology and toolsets while providing comprehensive and structured memory support and can pave the way for the next generation of agent development, shifting the focus from human-centered to AI-centered paradigms.

Abstract

Agent development kits (ADKs) provide effective platforms and tooling for constructing agents, and their designs are critical to the constructed agents' performance, especially the functionality for agent topology, tools, and memory. However, current ADKs either lack sufficient functional support or rely on humans to manually design these components, limiting agents' generalizability and overall performance. We propose OpenSage, the first ADK that enables LLMs to automatically create agents with self-generated topology and toolsets while providing comprehensive and structured memory support. OpenSage offers effective functionality for agents to create and manage their own sub-agents and toolkits. It also features a hierarchical, graph-based memory system for efficient management and a specialized toolkit tailored to software engineering tasks. Extensive experiments across three state-of-the-art benchmarks with various backbone models demonstrate the advantages of OpenSage over existing ADKs. We also conduct rigorous ablation studies to demonstrate the effectiveness of our design for each component. We believe OpenSage can pave the way for the next generation of agent development, shifting the focus from human-centered to AI-centered paradigms.

OpenSage: Self-programming Agent Generation Engine

TL;DR

OpenSage is proposed, the first ADK that enables LLMs to automatically create agents with self-generated topology and toolsets while providing comprehensive and structured memory support and can pave the way for the next generation of agent development, shifting the focus from human-centered to AI-centered paradigms.

Abstract

Agent development kits (ADKs) provide effective platforms and tooling for constructing agents, and their designs are critical to the constructed agents' performance, especially the functionality for agent topology, tools, and memory. However, current ADKs either lack sufficient functional support or rely on humans to manually design these components, limiting agents' generalizability and overall performance. We propose OpenSage, the first ADK that enables LLMs to automatically create agents with self-generated topology and toolsets while providing comprehensive and structured memory support. OpenSage offers effective functionality for agents to create and manage their own sub-agents and toolkits. It also features a hierarchical, graph-based memory system for efficient management and a specialized toolkit tailored to software engineering tasks. Extensive experiments across three state-of-the-art benchmarks with various backbone models demonstrate the advantages of OpenSage over existing ADKs. We also conduct rigorous ablation studies to demonstrate the effectiveness of our design for each component. We believe OpenSage can pave the way for the next generation of agent development, shifting the focus from human-centered to AI-centered paradigms.
Paper Structure (28 sections, 5 figures, 4 tables)

This paper contains 28 sections, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Existing ADKs and Limitations
  3. Key Techniques
  4. Overview
  5. Self-generating Agent Topology
  6. Dynamic Tool Synthesis
  7. Hierarchical Memory Management
  8. Short-term memory.
  9. Evaluation
  10. OpenSage on SOTA Benchmarks
  11. Evaluation of Self-Generating Agent Structure
  12. Evaluation of Tooling System
  13. Evaluation of Memory Management
  14. Conclusion and Future Works
  15. More Related Work
  16. ...and 13 more sections

Figures (5)

  • Figure 1: SageAgent (via OpenSage) vs. SOTA agents and ADKs on three popular agentic benchmarks. "G.3" refers to Gemini 3.
  • Figure 2: Overview of OpenSage framework, consisting of three key components. First, we enable AI to create different topologies while managing them in a unified agent pool. We then propose a hierarchical tool structure, including tool-specific sandboxing and states, and asynchronous execution. We design graph-based short-term and long-term memory with a memory agent to interact with them.
  • Figure 3: Ablation analysis of SageAgent built with OpenSage framework on a 300-instance subset of CyberGym, evaluating the impact of agent topology (left) and tooling system (right).
  • Figure 4: Resolved rate (left) and cost (right) for agents built with OpenSage framework on Terminal-Bench 2.0 using Gemini 3 Pro, GPT-5 Mini, and a large-small collaboration setup (Gemini 3 Pro + GPT-5 Mini), compared against GPT-5.
  • Figure 5: Ablation analysis of SageAgent built with OpenSage framework, comparing different memory designs (agentic, Mem0${}^g$, no memory) on SWE-Bench Pro.