Model Council Pro

将同一问题通过 OpenRouter 并行发送给 3 个及以上大语言模型，再由裁判模型比较各自回答，输出胜出模型、判断理由与综合后的最佳答案，并列出各模型及裁判的费用。适用于重要决策、代码评审、研究核对等需要多视角验证的场景。

aiwithabidi

热度 63.98

最近更新 2026-05-10 17:37:54

分类：知识依赖：内部调用大模型 / 第三方 API / 代码执行应用场景：研究 / 代码审查 / 调试排错 / 摘要总结 ...

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add model-council-pro

Metadata

name: model-council
description: Multi-model consensus system — send a query to 3+ different LLMs via OpenRouter simultaneously, then a judge model evaluates all responses and produces a winner, reasoning, and synthesized best answer. Like having a board of AI advisors. Use for important decisions, code review, research verification.
homepage: https://www.agxntsix.ai
license: MIT
compatibility: Python 3.10+, OpenRouter API key
metadata: {"openclaw":{"emoji":"🏛️","requires":{"env":["OPENROUTER_API_KEY"]},"primaryEnv":"OPENROUTER_API_KEY","homepage":"https://www.agxntsix.ai"}}

Model Council 🏛️

Get consensus from multiple AI models on any question.

Send your query to 3+ different LLMs simultaneously via OpenRouter. A judge model evaluates all responses and produces a winner, reasoning, and synthesized best answer.

When to Use

Important decisions — Don't trust one model's opinion
Code review — Get multiple perspectives on architecture choices
Research verification — Cross-check facts across models
Creative work — Compare writing styles and pick the best
Debugging — When one model is stuck, others might see the issue

How It Works

Your Question
    ├──→ Claude Sonnet 4    ──→ Response A
    ├──→ GPT-4o             ──→ Response B
    └──→ Gemini 2.0 Flash   ──→ Response C
                                    │
                              Judge (Opus) evaluates all
                                    │
                              ├── Winner + Reasoning
                              ├── Synthesized Best Answer
                              └── Cost Breakdown

Quick Start

# Basic usage
python3 {baseDir}/scripts/model_council.py "What's the best database for a real-time analytics dashboard?"

# Custom models
python3 {baseDir}/scripts/model_council.py --models "anthropic/claude-sonnet-4,openai/gpt-4o,google/gemini-2.5-pro" "Your question"

# Custom judge
python3 {baseDir}/scripts/model_council.py --judge "openai/gpt-4o" "Your question"

# JSON output
python3 {baseDir}/scripts/model_council.py --json "Your question"

# Set max tokens per response
python3 {baseDir}/scripts/model_council.py --max-tokens 2000 "Your question"

Configuration

Flag	Default	Description
`--models`	claude-sonnet-4, gpt-4o, gemini-2.0-flash	Comma-separated model list
`--judge`	anthropic/claude-opus-4-6	Judge model
`--max-tokens`	1024	Max tokens per council member
`--json`	false	Output as JSON
`--timeout`	60	Timeout per model (seconds)

Environment

Requires OPENROUTER_API_KEY environment variable.

Output Example

═══ MODEL COUNCIL RESULTS ═══

Question: What's the best way to handle auth in a microservices architecture?

── Council Member Responses ──

🤖 anthropic/claude-sonnet-4 ($0.0043)
Use a centralized auth service with JWT tokens...

🤖 openai/gpt-4o ($0.0038)
Implement OAuth 2.0 with an API gateway...

🤖 google/gemini-2.0-flash-001 ($0.0012)
Consider using service mesh with mTLS...

── Judge Verdict (anthropic/claude-opus-4-6, $0.0125) ──

🏆 Winner: anthropic/claude-sonnet-4
Reasoning: Most comprehensive and practical approach...

📝 Synthesized Answer:
The best approach combines elements from all three...

💰 Total Cost: $0.0218

Credits

Built by M. Abidi | agxntsix.ai YouTube | GitHub Part of the AgxntSix Skill Suite for OpenClaw agents.

📅 Need help setting up OpenClaw for your business? Book a free consultation