#OpenSage#self‑programming agents#agent development kit#LLM autonomy#topology generation#toolset creation#graph‑based memory#structured memory#software engineering toolkit#benchmark experiments#ablation study#AI‑centered development
📌 Key Takeaways
OpenSage enables LLMs to autonomously generate agents with self‑defined topology and toolsets.
The kit includes a hierarchical, graph‑based memory system for efficient structured memory management.
A specialized toolkit assists in software engineering tasks, distinguishing it from other ADKs.
Experimental evaluation on three benchmarks shows OpenSage outperforms existing agent development kits.
Ablation studies confirm the importance of each design component.
The work promotes a paradigm shift from human‑to AI‑centric agent creation.
📖 Full Retelling
WHO: The research team led by Hongwei Li, together with Zhun Wang, Qinrun Dai, Yuzhou Nie, Jinjun Peng, Ruitong Liu, Jingyang Zhang, Kaijie Zhu, Jingxuan He, Lun Wang, Yangruibo Ding, Yueqi Chen, Wenbo Guo, and Dawn Song.
WHAT: They present "OpenSage", the first agent development kit (ADK) that allows large language models (LLMs) to automatically generate agents, including their own topologies, toolsets, and a hierarchical, graph‑based memory system.
WHERE: The paper is hosted on arXiv, an open-access preprint repository.
WHEN: It was submitted on 18 February 2026.
WHY: Current ADKs lack comprehensive functional support or require manual human design, which restricts agent generality and performance. OpenSage seeks to shift from human‑centered to AI‑centered agent development by enabling LLMs to autonomously create and manage sophisticated agents.
The authors demonstrate the capabilities of OpenSage across three state‑of‑the‑art benchmarks, perform ablation studies to isolate the contribution of each component, and highlight how the new toolkit can accelerate future research in automated agent generation.
🏷️ Themes
Artificial Intelligence, Agent Development, Large Language Models, Automated Tool Generation, Hierarchical Memory Systems, Software Engineering, Benchmarking & Evaluation, Ablation Analysis
Entity Intersection Graph
No entity connections available yet for this article.
Deep Analysis
Why It Matters
OpenSage automates agent creation, reducing human design effort and improving performance across AI tasks. This shift to AI-centered development could accelerate deployment of autonomous systems.
Context & Background
Traditional agent development kits require manual design of topology and tools.
Existing ADKs lack comprehensive memory support.
OpenSage introduces self-programming agents with hierarchical memory and tool generation.
What Happens Next
Future work may integrate OpenSage with larger language models and real-world robotics. Researchers will likely benchmark it against more diverse tasks and explore its security implications.
Frequently Asked Questions
What is the main innovation of OpenSage?
It lets large language models automatically generate agent topologies, toolkits, and structured memory, eliminating manual design.
Which benchmarks were used to evaluate OpenSage?
The authors tested it on three state-of-the-art benchmarks covering software engineering and general agent tasks.
Is the code for OpenSage publicly available?
Yes, the paper provides links to code repositories and the authors encourage community contributions.
Original Source
--> Computer Science > Artificial Intelligence arXiv:2602.16891 [Submitted on 18 Feb 2026] Title: OpenSage: Self-programming Agent Generation Engine Authors: 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 View a PDF of the paper titled OpenSage: Self-programming Agent Generation Engine, by Hongwei Li and 13 other authors View PDF HTML Abstract: Agent development kits 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. Subjects: Artificial Intelligence (cs.AI) ; Cryptography and Security (cs.CR); Software Engineering (cs.SE) Cite as: arXiv:2602.16891 [cs.AI] (or arXiv:2602.16891v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2602.16891 Focus to learn more arXi...
