Talk2X -- An Open-Source Toolkit Facilitating Deployment of LLM-Powered Chatbots on the Web

TL;DR

Talk2X presents an open-source function-calling agent that enables efficient deployment of LLM-powered chatbots on websites using an adapted RAG framework backed by a vector database. By pre-building two collections (website content and asset metadata) and enabling iterative function calls, Talk2X reduces repeated crawling, lowers context size, and improves energy efficiency. Two usability studies show Talk2X outperforms standard website interaction in task completion time, accuracy, and perceived usability, while providing sources to audit AI outputs. The work offers a generalizable, scalable toolkit for web hosts to deploy transparent, auditable, and efficient chatbots, along with deployment guidance and identified limitations for future work.

Abstract

Integrated into websites, LLM-powered chatbots offer alternative means of navigation and information retrieval, leading to a shift in how users access information on the web. Yet, predominantly closed-sourced solutions limit proliferation among web hosts and suffer from a lack of transparency with regard to implementation details and energy efficiency. In this work, we propose our openly available agent Talk2X leveraging an adapted retrieval-augmented generation approach (RAG) combined with an automatically generated vector database, benefiting energy efficiency. Talk2X's architecture is generalizable to arbitrary websites offering developers a ready to use tool for integration. Using a mixed-methods approach, we evaluated Talk2X's usability by tasking users to acquire specific assets from an open science repository. Talk2X significantly improved task completion time, correctness, and user experience supporting users in quickly pinpointing specific information as compared to standard user-website interaction. Our findings contribute technical advancements to an ongoing paradigm shift of how we access information on the web.

Figures (10)

• Figure 1: Overview over the interactions and system architecture of Talk2X.
• Figure 2: The agents internal interaction flow upon receiving a user query.
• Figure 3: Building the website collection (a) and the asset collection (b).
• Figure 4: The user interface of Talk2AIoD depicting a search mask and response field.
• Figure 5: The procedure of Study 1. The meta information and asset questions can be found in \ref{['tab:questions_n_tasks']}.