Joint Search

Joint Search is a collaborative, chat-based fact-checking and research platform designed to provide deep, context-aware information retrieval and synthesis. At its core, the project connects a modern React/Vite frontend with a robust Express.js backend (utilizing PostgreSQL for data persistence and MinIO for S3-compatible object storage) to deliver a highly interactive user experience. Users can upload documents to knowledge bases, organize them with file or folder tags, and dynamically reference them inside chat conversations. To answer complex queries, the system integrates a powerful multi-agent orchestration workflow built on the Band framework. This workflow delegates tasks in a structured sequence: The Research Orchestrator manages quality and initiates the pipeline. The Research Planner decomposes the main query into distinct subtopics. Specialist Researcher Agents execute targeted web searches using Perplexity API models. The Medior Agent collects these findings, guides a single-round debate among researchers to resolve discrepancies, and synthesizes a final draft.