LLM Council

Query multiple AI models simultaneously and get a synthesized consensus answer through live dashboard voting. LLM Council routes your prompt to Claude, GPT, Gemini, and other models in parallel, streams their responses in real time, and ranks them by anonymous peer evaluation.

What Is This?

• Multi-model orchestration that sends one prompt to several AI models at once and displays all responses side by side
• Live web dashboard with real-time streaming via Server-Sent Events, markdown rendering, and light/dark themes
• Anonymous voting phase where models rank each other's responses, then synthesizes a consensus answer

Why Use It?

• Catch errors and blind spots — single models can be confidently wrong; multiple models surface disagreements and edge cases
• Reduce bias — anonymous peer voting evaluates reasoning quality, not brand reputation
• Benchmark performance — compare latency, accuracy, and task-specific strengths across model versions
• Validate high-stakes answers — legal, medical, security, and financial queries benefit from multi-model validation before reaching users

How to Use It?

Configure your models and gateway credentials:

COUNCIL_MODELS = [
    {"id": "claude-3-5-sonnet", "provider": "anthropic"},
    {"id": "gpt-4o", "provider": "openai"},
    {"id": "gemini-1.5-pro", "provider": "google"},
]

GATEWAY_BASE_URL = "https://your-gateway-endpoint.example.com"
GATEWAY_API_KEY = "your_api_key_here"

Launch the server:

from http.server import ThreadingHTTPServer
from council_handler import CouncilRequestHandler

server = ThreadingHTTPServer(("localhost", 8080), CouncilRequestHandler)
print("Council dashboard running at http://localhost:8080")
server.serve_forever()

Submit a prompt through the dashboard. All models receive it simultaneously, stream responses in parallel, then enter voting phase. The dashboard aggregates rankings and highlights the consensus winner.

When to Use It?

• Legal interpretation, medical summaries, security analysis, or financial reasoning where wrong answers carry high cost
• Prompt engineering — test how different models interpret ambiguous phrasing before committing to single-model deployment
• Research — survey how current models handle a specific domain in one parallel session instead of sequential experiments

Important Notes

• API costs multiply per model added; a three-model council triples token spend per query
• SSE streams require at least 120 seconds for long-form responses; configure firewalls and reverse proxies accordingly
• If one model times out, voting proceeds with remaining responses and marks the timed-out model unavailable

Try It in Happycapy

1. Open Happycapy in your browser — no install or signup needed.
2. Describe what you want in one sentence. Example: "Compare how Claude, GPT-4, and Gemini explain quantum entanglement to a high school student."
3. Watch all three models respond in real time on the live dashboard, then see the consensus ranking and synthesized answer.