ARGUS — Argus Debate AI

A debate-native, multi-agent AI framework for evidence-based reasoning with structured argumentation, decision-theoretic planning, and full provenance tracking. ARGUS implements the Research Debate Chain (RDC) — a novel approach that structures knowledge evaluation as multi-agent debates. Instead of single-pass LLM inference, ARGUS orchestrates Moderator, Specialist, Refuter, and Jury agents that gather evidence, generate rebuttals, and render verdicts through Bayesian aggregation. Key innovations include: (1) Conceptual Debate Graph (C-DAG) — a directed graph where propositions, evidence, and rebuttals are nodes with signed edges representing support/attack relationships; (2) Evidence-Directed Debate Orchestration (EDDO) — an algorithm managing multi-round debates with convergence detection; (3) Value of Information Planning using Expected Information Gain (EIG); (4) Full PROV-O compatible provenance tracking with SHA-256 hash-chain integrity. Supports 27+ LLM providers, 50+ tools, hybrid BM25+FAISS retrieval, and ships with a TUI, Streamlit sandbox, and REST API. Available on PyPI as argus-debate-ai.