---
title: "ATO vs Karpathy llm-council"
canonical: "https://agentictool.ai/compare/llm-council"
description: "llm-council popularized multi-model deliberation. ATO is the maintained, multi-provider, tool-calling version with a local audit trail and a full agent ops product around the same premise."
---

# ATO vs Karpathy's llm-council

Same premise (ask more than one model). Different product: tools, providers, audit, and ongoing maintenance.

| Dimension | ATO | llm-council |
|---|---|---|
| Status | Maintained product (desktop + CLI + MCP) | Popular demo; author marked it unsupported |
| Providers | Native CLI runtimes + many API providers | Typically OpenRouter-shaped setups |
| Tool calling | Models can read_file / grep / git_log in your repo | Text debate without repo tools by default |
| Audit trail | Per-dispatch receipts in local SQLite | Session chat history in the app |
| Beyond debate | Replay, regression, cost routing, agent files, Pro automation | Council primitive only |
| License | MIT | MIT |

## When to choose ATO

- You want councils that verify claims against a real codebase
- You need multi-provider auth without an OpenRouter lock-in
- You want receipts, replay, and ops around the deliberation

## When to choose the other tool(s)

- You want the smallest possible demo of multi-model debate
- You are studying the original Karpathy design specifically

## Using them together

Read the original idea, then run the maintained shape with tools: brew install willnigri/ato/ato && ato demo-war-room

## Further reading

- [Blog: rebuilt llm-council with tools](/posts/llm-council-tool-calls.html)
- [Coordination benchmark](/posts/does-multi-model-coordination-beat-a-single-model)
- [FAQ](/faq)