Build Your Personalized Research Group: A Multiagent Framework for Continual and Interactive Science Automation
Ed Li, Junyu Ren, Xintian Pan, Cat Yan, Chuanhao Li, Dirk Bergemann, Zhuoran Yang
2025-10-20
Summary
This paper introduces a new AI system called freephdlabor designed to automate scientific research, going beyond what current AI can do in science.
What's the problem?
Current AI systems trying to do science have trouble because they follow a set plan and can't easily change course when they find something unexpected, and they also struggle to keep track of all the information they gather over a long research project, making it hard to build on previous work.
What's the solution?
The researchers created freephdlabor, which is like a team of AI 'agents' that can dynamically decide what to do next based on their findings, and it's built in a way that allows scientists to easily customize it for their specific research needs. It also has systems to automatically organize information, allow the agents to communicate effectively, remember what they've learned, and even allow humans to step in and guide the process when needed.
Why it matters?
This work is important because it moves automated science from being a series of one-off experiments to a continuous research process that learns and improves over time, potentially speeding up discoveries and allowing scientists to focus on the bigger picture. It also makes this kind of automated research more accessible to scientists in different fields.
Abstract
The automation of scientific discovery represents a critical milestone in Artificial Intelligence (AI) research. However, existing agentic systems for science suffer from two fundamental limitations: rigid, pre-programmed workflows that cannot adapt to intermediate findings, and inadequate context management that hinders long-horizon research. We present freephdlabor, an open-source multiagent framework featuring fully dynamic workflows determined by real-time agent reasoning and a \textit{modular architecture} enabling seamless customization -- users can modify, add, or remove agents to address domain-specific requirements. The framework provides comprehensive infrastructure including automatic context compaction, workspace-based communication to prevent information degradation, memory persistence across sessions, and non-blocking human intervention mechanisms. These features collectively transform automated research from isolated, single-run attempts into continual research programs that build systematically on prior explorations and incorporate human feedback. By providing both the architectural principles and practical implementation for building customizable co-scientist systems, this work aims to facilitate broader adoption of automated research across scientific domains, enabling practitioners to deploy interactive multiagent systems that autonomously conduct end-to-end research -- from ideation through experimentation to publication-ready manuscripts.