ReasoningBank
Provides AI agents with memory-enhanced reasoning capabilities through the Model Context Protocol.
About
ReasoningBank is a memory-enhanced reasoning system implemented based on the 'ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory' paper. It empowers AI agents by offering robust experiential memory management via the Model Context Protocol (MCP). The system automatically extracts reasoning experiences from both successful and failed trajectories, providing intelligent retrieval mechanisms to enhance an agent's problem-solving abilities and decision-making processes.
Key Features
- Compatible with various large language models (LLMs) such as DashScope, OpenAI, and Claude.
- Features a plug-in architecture for easy extension of new retrieval strategies and storage backends.
- Automatically extracts reasoning experiences from successful and failed task trajectories.
- 1 GitHub stars
- Supports multiple retrieval strategies, including cosine similarity and a hybrid scoring model.
- Offers asynchronous memory extraction to prevent blocking AI agent operations.
Use Cases
- Allowing AI agents to automatically extract and save new reasoning experiences after completing tasks.
- Enhancing AI agent performance by providing access to learned historical experiences.
- Enabling AI agents to retrieve relevant past memories before tackling new tasks.
Sponsored