consensus-pipeline

🧠 Consensus Pipeline

📖 中文文档 · ⚡ Quick Start · 🏗️ Architecture · 📖 Usage · 📦 Releases · 🔗 Related: Multivest Multi-agent debate framework for academic research. Instead of one AI writing a literature review for you — an AI team interviews you, debates each claim across methodological perspectives, reaches consensus via dynamic mathematical convergence, and verifies every citation against source abstracts and full texts. ⚡ Quick Start

Pick the fastest way to get running (see 📖 Usage for full configuration, full-text PDF breakpoint, and custom endpoints):

🚀 1. One-shot installer (Clones + Installs Deps + Prints MCP Config)

🤖 2. DSH Plugin (AI-Driven with Floating Control Panel)

Tell DSH "consensus-pipeline start" (or "共识管线开始需求调研") to launch the interview and debate pipeline.

🖥️ 3. Local Streamlit Web UI / Headless CLI ❓ Why Not Just Ask ChatGPT or Elicit?

A single LLM produces confident-sounding answers with no cross-validation — hallucinations slip through, conflicting perspectives get flattened, and you cannot tell which conclusions are robust vs. speculative.

Consensus Pipeline replaces one-shot generation with structured multi-agent debate as a quality gate: every claim is challenged by independent "departments," contradictions are surfaced explicitly, and final conclusions carry per-claim confidence annotations (e.g., "42/77 papers, high confidence"). 📸 What It Looks Like Step 1: Requirement Interview The pipeline starts by interviewing you — an AI agent asks clarifying questions to understand your research scope, constraints, and goals.

Step 2: Smart Department Configuration Based on your topic, the AI auto-generates 10+ specialized debate departments with multiple debaters per department. Each debater argues from a distinct methodological perspective.

Step 3: Multi-Round Debate with Dynamic Convergence Watch debaters argue in real time. Each round, debaters present positions, challenge assumptions, and refine arguments. The pipeline runs 3–8 rounds per department, stopping early once debaters converge via stance coefficient of variation ($CV$) and Kendall's $W$.