Le Répertoire des AgentsFR · EN · ES · JA · HI

“I want a solid second opinion before a big decision”

Executives, investors, freelancers and anyone who makes costly decisions based on a single AI's answers. If one model has ever answered you confidently… and wrong, this profile is for you.

Recommended agent: LLM Council

💰 Free software (public repository, no open-source license); you pay for the model calls (about 2 to 3 times the cost of asking a single model). Level: Advanced 🌍 United States Free software; no subscription — only the models' API calls are billed by usage

Price verified on 2026-10-29.

LLM Council, a public project by Andrej Karpathy (the repo is visible, but has no license file: not open source in the legal sense): you ask your question to several models from different families (GPT, Claude, Gemini, Grok…), each answers on its own, then they read and rank each other anonymously, and a “chairman” model synthesizes the final verdict.

How it works

Trigger: You submit an important question to the council — queries each model in parallel, anonymizes the answers, organizes the cross-review and has a chair synthesize, and you get a debated answer, with the disagreements visible instead of hidden.

Step-by-step setup

Grab the public project on GitHub (github.com/karpathy/llm-council) and install it locally.
Create an OpenRouter account and add a small credit to pay for the model calls.
Choose the council's models (four models from different families) and the chairman model.
Ask a real decision question and compare the verdict with a single model's answer.

⚠️ Guardrails: A council of models is still a group of models: they can share the same blind spot. Keep irreversible decisions (legal, health, large amounts) for real human experts, and don't submit confidential data without checking where it flows.

🔧 Technical details

Installation: Local application
Prerequisites: Python 3.10+ with uv for the backend (FastAPI) and Node.js/npm for the frontend (React + Vite), plus an OpenRouter API key with credits.
Hardware: non_applicable
Self-hosting: yes — local web application launched from the repo (start.sh script, or backend/frontend started separately), accessible at localhost:5173.
License: the official repo contains no license file (no LICENSE at the root as of 2026-11-01).
Models: Models accessible only via OpenRouter, configured in backend/config.py (by default: GPT-5.1, Gemini 3 Pro, Claude Sonnet 4.5, Grok 4, with a “Chairman” model that synthesizes). Personal OpenRouter key required; local models not documented.
Data: Conversations are stored locally as JSON files in data/conversations/; requests are sent to OpenRouter, which routes them to the providers of the council's models.
Operations: an avowed demo project (“vibe coded”, with no support promised by the author in the README).

Not documented in the official sources: API & integrations, Documented limits.

Official sources: github.com — verified on November 1, 2026.

Find your agent with the quiz →

Frequently asked questions

How much does LLM Council cost?

Free software (public repository, no open-source license); you pay for the model calls (around 2 to 3 times the cost of asking a single model). Free software; no subscription, only the models' API calls are billed by usage. Price verified on October 29, 2026.

Is LLM Council free?

Yes. Free software; no subscription, only the models' API calls are billed by usage.

Is LLM Council hard to get started with?

Level: Advanced. Setup takes 4 steps; the first: Grab the public project on GitHub (github.com/karpathy/llm-council) and install it locally.

What limits and guardrails should I know about with LLM Council?

A council of models is still a group of models: they can share the same blind spot. Keep irreversible decisions (legal, health, large amounts) for real human experts, and don't submit confidential data without checking where it flows.