CoSearchAgent: A Lightweight Collaborative Search Agent with Large Language Models

TL;DR

The paper addresses the challenge of enabling lightweight, real-time collaborative search in multi-user chat environments by introducing CoSearchAgent, a Slack plugin powered by large language models. It presents an end-to-end architecture with query rewriting, clarifying-question generation, retrieval-augmented generation, and grounding in web-derived references, all delivered within a single-round interaction to support discussion in multi-party conversations. Key contributions include an end-to-end workflow that extracts and cites references from search results, an open-source Slack plugin with logging for behavioral analysis, and a case study comparing direct search, Wizard of Oz, and the proposed CoSearchAgent approach. The work demonstrates improved search efficiency and grounded answer quality in collaborative contexts, offering a practical platform for advancing research on collaborative information seeking in instant messaging ecosystems.

Abstract

Collaborative search supports multiple users working together to accomplish a specific search task. Research has found that designing lightweight collaborative search plugins within instant messaging platforms aligns better with users' collaborative habits. However, due to the complexity of multi-user interaction scenarios, it is challenging to implement a fully functioning lightweight collaborative search system. Therefore, previous studies on lightweight collaborative search had to rely on the Wizard of Oz paradigm. In recent years, large language models (LLMs) have been demonstrated to interact naturally with users and achieve complex information-seeking tasks through LLM-based agents. Hence, to better support the research in collaborative search, in this demo, we propose CoSearchAgent, a lightweight collaborative search agent powered by LLMs. CoSearchAgent is designed as a Slack plugin that can support collaborative search during multi-party conversations on this platform. Equipped with the capacity to understand the queries and context in multi-user conversations and the ability to search the Web for relevant information via APIs, CoSearchAgent can respond to user queries with answers grounded on the relevant search results. It can also ask clarifying questions when the information needs are unclear. The proposed CoSearchAgent is highly flexible and would be useful for supporting further research on collaborative search. The code and demo video are accessible.

Figures (3)

• Figure 1: An instance of lightweight collaborative search. The collaborative search system needs to understand the conversational context of interactions between two users and propose search results to the users.
• Figure 2: The overall architecture of CoSearchAgent. Given a context of multi-party dialogue and a query posed by a user, CoSearchAgent provides its response through the following three steps: (I) Query Processing: Rewriting the query based on the dialogue context and asking clarification questions for ambiguous parts of the query; (II) Search Results Presentation: Retrieve search results, extract relevant contents related to the query, and provide them to users as references; (III) Retrieval-Augmented Generation: Responding to the user's query relying on the generated references.
• Figure 3: Example of CoSearchAgent's usage in Slack. Similar to the Wizard of Oz approach, CoSearchAgent can rewrite the query accurately for searching, and generate the accurate answer based on search results for easier user reading.