Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/karpathy/llm-council/llms.txt

Use this file to discover all available pages before exploring further.

LLM Council talks to every council member — OpenAI, Anthropic, Google, xAI, Meta, and more — through a single API endpoint provided by OpenRouter. Instead of creating and managing separate API keys for each AI provider, you purchase credits on OpenRouter and use one key for everything. This guide walks through account creation, key configuration, and how the key flows through the application.

​
What OpenRouter provides

OpenRouter is a unified API gateway that normalizes hundreds of LLMs behind a single POST https://openrouter.ai/api/v1/chat/completions endpoint. Every request you make to LLM Council — council member queries in Stage 1 and Stage 2, the Chairman call in Stage 3, and background title generation — goes to this one endpoint with an Authorization: Bearer <your-key> header. OpenRouter handles routing to the underlying provider, billing, and rate limiting transparently.

​
Setting up your account and key

1
Sign up at OpenRouter
2
Go to openrouter.ai and create a free account. No model-specific agreements are needed — OpenRouter handles provider terms on your behalf.
3
Purchase credits or enable auto top-up
4
Navigate to the Credits section of your dashboard. You can add a one-time credit balance or enable automatic top-up so your council never gets interrupted mid-conversation. Frontier models (GPT-5, Claude Opus, Gemini Pro) cost more per call than mid-tier models; the model list shows per-token pricing for every option.
5
Create an API key
6
Go to the Keys section and click Create Key. Give it a descriptive name (e.g., llm-council-local). The key will be displayed once — copy it now. It begins with sk-or-v1-.
7
Create a .env file in the project root
8
In the root of the llm-council repository, create a file named .env with the following content:
9
OPENROUTER_API_KEY=sk-or-v1-your-key-here
10
Replace sk-or-v1-your-key-here with the key you copied in the previous step.
11
The .env file is listed in .gitignore and will not be committed to version control. Never paste your API key directly into config.py or any other tracked file.
12
Start (or restart) the backend
13
The key is loaded at startup via python-dotenv in backend/config.py:
14
from dotenv import load_dotenv
load_dotenv()

OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY")
15
If the backend is already running, stop it and restart it so it picks up the new environment variable:
16
uv run python -m backend.main

​
Credit consumption per message

Every user message triggers a fixed number of model calls. With N council members:
StageCallsDescription
Stage 1NCouncil members answer the question in parallel
Stage 2NCouncil members rank each other’s answers in parallel
Stage 31Chairman synthesizes the final response
Title generation1google/gemini-2.5-flash generates a 3–5 word title (first message only)
Total2N + 2Per message (2N + 1 for follow-up messages)
With the default 4-model council, that is 10 calls on the first message and 9 calls on every subsequent message in the same conversation.
You can verify your key is active and check your remaining credits at any time from the OpenRouter dashboard, or by calling the OpenRouter keys API directly. The dashboard also shows a per-model breakdown of your usage so you can see exactly which council members are consuming the most credits.
If OPENROUTER_API_KEY is missing or invalid, every model query will fail with an HTTP 401 authentication error. Stage 1 will return zero responses, and LLM Council will return: "All models failed to respond. Please try again." — check the backend terminal output for the raw error detail if this happens.

Build docs developers (and LLMs) love

Get started for freeTalk to us