ReasonGraph: Visualisation of Reasoning Paths

TL;DR

ReasonGraph tackles the challenge of understanding LLM reasoning by providing a unified, real-time visualization platform. It supports six reasoning methods and 50+ models across major providers, with Mermaid-based diagrams and a modular, API-driven backend. Key contributions include a unified visualization platform, extensible framework, and multi-domain applications, demonstrated with high parsing reliability, fast visualization, and good usability. The tool is open-source, enabling broader accessibility, reproducibility, and development of LLM-based reasoning tooling.

Abstract

Large Language Models (LLMs) reasoning processes are challenging to analyze due to their complexity and the lack of organized visualization tools. We present ReasonGraph, a web-based platform for visualizing and analyzing LLM reasoning processes. It supports both sequential and tree-based reasoning methods while integrating with major LLM providers and over fifty state-of-the-art models. ReasonGraph incorporates an intuitive UI with meta reasoning method selection, configurable visualization parameters, and a modular framework that facilitates efficient extension. Our evaluation shows high parsing reliability, efficient processing, and strong usability across various downstream applications. By providing a unified visualization framework, ReasonGraph reduces cognitive load in analyzing complex reasoning paths, improves error detection in logical processes, and enables more effective development of LLM-based applications. The platform is open-source, promoting accessibility and reproducibility in LLM reasoning analysis.

Figures (4)

• Figure 1: The ReasonGraph UI with a query input header and dual-panel layout: the left panel shows Reasoning Settings for configuring model API parameters along with Raw Model Output displaying the unprocessed text response from the LLM. The right panel contains Visualization Settings for adjusting diagram parameters and Visualization Results showing a flowchart illustration of the reasoning process.
• Figure 2: Comparison between plain text (bottom) and organized tree visualization (top) for the same reasoning process using beam search method. The blue box is the initial question, the darker blue box highlights the selected reasoning path, and the final solution is shown in a green box.
• Figure 3: Visualization examples of four sequential reasoning methods: Chain-of-Thoughts (top-left), Self-refine (top-center), Least-to-Most (top-right), and Self-consistency (bottom-left). In Self-refine, yellow boxes indicate reflection and improvement steps; in Least-to-Most, light blue boxes are original and decomposed questions while green boxes show intermediate and final answers.
• Figure 4: The framework of ReasonGraph, consisting of four main layers: UI Components for user involvement, Client-side for frontend processing, RESTful Routes for API bridge, and a modular backend comprising Configuration Manager, API Factory for LLM integration, and Reasoning Methods implementation.