ReasonGraph: An Open-Source Project for Visualizing LLM Reasoning 193

What is ReasonGraph?

ReasonGraph is an open-source project for visualizing and analyzing the reasoning processes of LLMs. It supports both sequential reasoning and tree-based reasoning methods, and integrates with more than 50 advanced models from major LLM providers including Anthropic, Openai, Google, and Together.AI. It turns the AI's thinking process into a visual graph, supports real-time updates and interactive analysis, and the resulting graphs can be exported as SVG with a single click.

Key Features of ReasonGraph

• Unified Visualization Framework: ReasonGraph provides real-time graph rendering and analysis, supporting comparisons across multiple reasoning methods.

• Modular and Extensible Design: Adopts a standardized API that facilitates the integration of new reasoning methods and models. Model Support: Integrates more than 50 state-of-the-art models from major LLM providers such as Anthropic, OpenAI, Google, and Together.AI.

• Reasoning Methods: Supports sequential reasoning (such as Chain-of-Thoughts, Self-refine, etc.) and tree-based reasoning (such as Beam Search, Tree-of-Thoughts, etc.).

• Modular Framework: Standardized API that facilitates the integration of new reasoning methods and models.

• User-Friendly Interface: Provides intuitive components, real-time reasoning path visualization, configurable parameter settings, and selection of meta-reasoning methods.

• Meta-Reasoning Capability: Allows the model to select the most appropriate reasoning method on its own.

Visualization Methods of ReasonGraph

• Sequential Reasoning Methods: Includes methods such as Chain-of-Thoughts, Self-refine, Least-to-Most, and Self-consistency, with reasoning steps displayed through a directed graph layout.

• Tree-Based Reasoning Methods: Such as Beam Search and Tree-of-Thoughts, with reasoning paths displayed through a tree structure.

Architecture of ReasonGraph

ReasonGraph adopts a modular framework, with the frontend handling visualization logic and user interaction, and the backend using Flask to implement configuration management, LLM integration, and reasoning method encapsulation. The platform supports dynamic switching of reasoning methods and models without modifying the framework.

Use Cases for ReasonGraph

• Academic Research: Supports comparative studies of different reasoning methods and evaluation of model capabilities.

• Education: Used for teaching principles of logical reasoning and demonstrating LLM decision-making processes.

• Development: Helps optimize prompt engineering and select the optimal reasoning method.

GitHub Repository:https://github.com/ZongqianLi/ReasonGraph

Paper Link:https://arxiv.org/html/2503.03979v1

Project Homepage:https://powerdrill.ai/discover/summary-reasongraph-visualisation-of-reasoning-paths-cm7z9j3oqk9wh07rs5ho63mee