OpenManus & OpenManus-RL
Published:
OpenManus-RL is the follow-up project of OpenManus. I collaborated with the official OpenManus team as a core contributor to OpenManus-RL in an open-source ecosystem with 60,000+ GitHub stars.
Key contributions include:
- Implemented a ReAct-style reasoning architecture with Think, Act, and Tool-use stages.
- Built multi-agent collaboration workflows.
- Built an MCP-based tool-library framework for automated tool invocation and output processing.
- Constructed high-quality SFT datasets for agent benchmarks including AlfWorld and WebShop.
- Designed step-level reward signals to improve agent planning and memory capabilities.
