Coinvisor: An RL-Enhanced Chatbot Agent for Interactive Cryptocurrency Investment Analysis

TL;DR

Coinvisor introduces a reinforcement learning–enhanced chatbot that orchestrates multi-step tool usage across specialized data analytics tools and domain-specific report agents to deliver real-time, multidimensional cryptocurrency investment analysis. By formulating tool selection as a finite-horizon MDP and optimizing with PPO, it achieves substantial improvements in tool-calling recall and F1 score while maintaining high precision. In user studies with 20 investors, Coinvisor achieves strong satisfaction (4.64/5) and is preferred over general LLMs and crypto platforms (4.62/5), highlighting its practical impact on reducing effort and improving decision quality. The system emphasizes safety by design, avoiding autonomous transaction execution and instead delegating final actions to the user, with future work aimed at secure, automated on-chain interactions.

Abstract

The cryptocurrency market offers significant investment opportunities but faces challenges including high volatility and fragmented information. Data integration and analysis are essential for informed investment decisions. Currently, investors use three main approaches: (1) Manual analysis across various sources, which depends heavily on individual experience and is time-consuming and prone to bias; (2) Data aggregation platforms-limited in functionality and depth of analysis; (3) Large language model agents-based on static pretrained models, lacking real-time data integration and multi-step reasoning capabilities. To address these limitations, we present Coinvisor, a reinforcement learning-based chatbot that provides comprehensive analytical support for cryptocurrency investment through a multi-agent framework. Coinvisor integrates diverse analytical capabilities through specialized tools. Its key innovation is a reinforcement learning-based tool selection mechanism that enables multi-step planning and flexible integration of diverse data sources. This design supports real-time interaction and adaptive analysis of dynamic content, delivering accurate and actionable investment insights. We evaluated Coinvisor through automated benchmarks on tool calling accuracy and user studies with 20 cryptocurrency investors using our interface. Results show that Coinvisor improves recall by 40.7% and F1 score by 26.6% over the base model in tool orchestration. User studies show high satisfaction (4.64/5), with participants preferring Coinvisor to both general LLMs and existing crypto platforms (4.62/5).

Figures (7)

• Figure 1: Comparison of three approaches to cryptocurrency investment queries: (left) standard LLM relying on pre-trained knowledge, (middle) tool-augmented LLM lacking multi-dimensional analysis capability with minimal tool invocation, and (right) multi-agent approach with coordinated multi-tool orchestration for comprehensive multidimensional analysis.
• Figure 2: The overview of Coinvisor.
• Figure 3: Examples of different trajectories during model training. The "$\rightarrow$" represents a high-reward trajectory, where the model generates a comprehensive sequence of tool calls. In contrast, the "$\rightarrow$" represents a low-reward trajectory, involving only a minimal number of tool calls and limited information gathering.
• Figure 4: The process of conducting security analysis on the source code of smart contracts.
• Figure 5: The workflow of multiple agents integrated in Coinvisor.