A multi-model consensus system that queries multiple LLMs simultaneously and uses a high-tier judge model to synthesize the best possible answer.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install model-council
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install model-council using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Model Council?
Model Council is a sophisticated framework within the Openclaw Skills ecosystem designed to eliminate single-model bias by leveraging a board of AI advisors. By broadcasting a single query to three or more distinct large language models through the OpenRouter API, this tool ensures you receive a diverse range of technical perspectives and creative insights. It is particularly valuable for users who need to cross-verify information or find the most optimal solution among competing AI outputs.
Once the council members provide their responses, a designated judge model evaluates each entry to determine a winner based on reasoning and logic. The system then produces a synthesized final answer that captures the best elements from all contributors. This makes Model Council an essential asset for anyone looking to build high-confidence workflows using Openclaw Skills, providing a layer of validation that single-agent queries simply cannot match.
Model Council Use Cases
- Validating critical technical or business decisions by aggregating multiple AI perspectives.
- Performing deep code reviews to identify architecture patterns or potential security flaws from different model viewpoints.
- Verifying research facts and data points across multiple distinct training sets to ensure accuracy.
- Comparing creative writing styles or marketing copy to select the most effective tone and structure.
- Solving complex debugging issues where one model might be stuck but others can identify the root cause.
How Model Council Works
- The user submits a specific question or technical prompt to the Model Council script.
- The system simultaneously broadcasts the query to a set of specified LLMs (such as Claude, GPT-4o, and Gemini) via OpenRouter.
- Each council member generates an individual response along with its associated token cost and latency data.
- A designated high-tier judge model (like Claude 3.5 Sonnet or Opus) analyzes all council member responses for accuracy, depth, and utility.
- The judge selects a winner, explains the rationale for the choice, and creates a synthesized best answer.
- The final output is presented with the winning model, the detailed reasoning, and a total cost breakdown for the entire operation.
Model Council Setup
To use this within your Openclaw Skills environment, you must have an active OpenRouter API key and Python 3.10+ installed.
# Set your environment variable
export OPENROUTER_API_KEY='your_api_key_here'
# Run a basic consensus query
python3 scripts/model_council.py "What is the most secure way to store JWTs in a web application?"
# Run with custom models and judge
python3 scripts/model_council.py --models "anthropic/claude-3.5-sonnet,openai/gpt-4o" --judge "anthropic/claude-3-opus" "Explain quantum entanglement."
Model Council Data Schema & Taxonomy
The skill organizes its output into a structured format that highlights the comparison between models. Below is the data taxonomy used for the results:
| Attribute | Description |
|---|---|
| Question | The original query processed by the council |
| Council Responses | The raw text output from each individual model |
| Winner | The specific model identified by the judge as having the best answer |
| Reasoning | A technical justification for why the winner was selected |
| Synthesized Answer | A consolidated response combining the strengths of all council members |
| Total Cost | The aggregated API cost for all members and the judge |
When using the --json flag, this data is returned as a parseable object for integration with other Openclaw Skills.
Model Council Advanced Features
- Support for custom model arrays allowing users to compare specific versions of LLMs from different providers.
- Configurable judge models to balance between evaluation depth and API cost management.
- JSON output mode for seamless programmatic integration into automated developer pipelines.
- Adjustable token limits and timeout settings per council member for performance optimization.
- Full compatibility with the broader suite of Openclaw Skills for multi-agent research and development tasks.
SKILL.md
Loading
Related Openclaw Skills
Model Council
A multi-model consensus engine that queries several LLMs simultaneously and uses a judge model to synthesize the best possible answer.
Model Audit
A diagnostic tool within the Openclaw Skills ecosystem designed to audit LLM stacks against live OpenRouter pricing and performance benchmarks.
Model Audit
A comprehensive auditing tool that compares your current LLM stack against live OpenRouter pricing to find cheaper and more powerful model alternatives.
Mercury Banking
A professional-grade Mercury banking integration for AI agents that automates transaction tracking, cash flow analysis, and financial categorization.
Model Intel
A powerful tool for fetching live LLM pricing, capabilities, and performance comparisons directly from OpenRouter.
Model Intel
A real-time intelligence tool for comparing LLM costs, capabilities, and task-specific performance via the OpenRouter API.
Featured*
