consensus

consensus is a web application (+command line tool) to ask multiple models a question, and provide a synthesized consensus answer using one model as the judge.

Overview

ReactJS and Tailwind for the frontend, .NET 9 with Microsoft Agent Framework and Quartz for the backend/API.

The app builds prompts for building a consensus using prompt templates and then the first LLM as the judge.

Screenshot

Quickstart

docker run -d \
  -p 8080:8080 \
  -e Consensus__ApiEndpoint="https://openrouter.ai/api/v1" \
  -e Consensus__ApiKey="your-api-key-here" \
  -e Consensus__Models__0="openai/gpt-4" \
  -e Consensus__Models__1="anthropic/claude-3-opus" \
  -e Consensus__Models__2="microsoft/phi-4" \
  -e Consensus__Models__3="google/gemini-2.5-flash" \
  -v $(pwd)/output:/app/output \
  --name consensus \
  ghcr.io/yetanotherchris/consensus:latest

Now go to http://localhost:8585/

Or use Docker compose:

services:
  consensus:
    image: ghcr.io/yetanotherchris/consensus:latest
    container_name: consensus
    ports:
      - "8085:8080"
    environment:
      - Consensus__ApiEndpoint=https://openrouter.ai/api/v1
      - Consensus__ApiKey=your-api-key-here
      - Consensus__Models__0=openai/gpt-4
      - Consensus__Models__1=anthropic/claude-3-opus
      - Consensus__Models__2=google/gemini-pro
      - Consensus__Models__3=microsoft/phi-4
    volumes:
      - ./output:/app/output
    restart: unless-stopped

Example prompt and model costs

Local development

• Install .NET 9 and NodeJS
• Clone the repo
• Add environment variables:
  • CONSENSUS_API_ENDPOINT e.g. https://openrouter.ai/api/v1
  • CONSENSUS_API_KEY e.g. sk-blah-blah
• cd .\src\Consensus.Api\ and dotnet run
• (New terminal) cd .\src\Consensus.Web\ and npm run dev
• Go to http://localhost:5173/ in a browser (or the port Vite provides for npm run dev)