Reproduce “TaskGen: A Task-Based, Memory-Infused Agentic Framework using StrictJSON” 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
John Chong Min Tan, Prince Saroj, Bharat Runwal, Hardik Maheshwari, Brian Lim Yi Sheng, Richard Cottrill, Alankrit Chona, Ambuj Kumar and 1 more
TaskGen, an open-source agentic framework from Simbian AI and NUS, introduces a task-based, memory-infused architecture using StrictJSON for reliable and concise LLM-generated outputs. It achieves high performance across diverse tasks including dynamic maze navigation, web browsing, and complex mathematical problem-solving, demonstrating superior efficiency and robustness.
Key insights
- Prioritizing conciseness and structured output via `StrictJSON` significantly enhances LLM agent efficiency, reliability, and cost-effectiveness by reducing token usage and parsing errors.
- Hierarchical task decomposition and explicit mapping of subtasks to `Equipped Functions` enable robust problem-solving by constraining the action space and ensuring guaranteed executability.
- Intelligent, context-aware memory management, including a `Memory Bank` with RAG and "need-to-know" information sharing, is crucial for sustaining long-running, complex agent behaviors without cognitive overload.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.