Skip to the contentAutomation & AI agents
Ami StudioOleksii Zahlukha

Automation & AI agentsCouncil of Models — eight AI models arguing in one chat

Council of Models — eight AI models arguing in one chat

A group chat where eight models from different vendors address each other by name, argue, and produce one written conclusion.

How it's wired

InInitialiseCreate a runReply at onceConfigRegisterPreparing the openingHost: openingParse the openingSave the openingPrepare ALines ACollect ASave APause 1Preparing the directionHost: directionParse the directionSave the directionPrepare BLines BCollect BSave BPause 2Preparing the verdictHost: verdictParse the verdictSave the verdictDoneResult to TelegramPreparing the fact queryExtract the factsParse the queriesBrave searchCollect the factsResearch: queriesResearch: modelResearch: parseResearch: BraveResearch: factsKPI: collectKPI: writeKPI inKPI rowsKPI pageKPI outKey accepted?No access
Council debate — group chat48 · 46

Drawn from the live workflow — every node and every arrow is really there. The labels are translated from the originals.

What it is

A single language model has no way of noticing that it is wrong: confidence in an answer has nothing to do with its correctness. The simplest known fix is to ask several and compare. Here that is arranged as a chat: eight models see each other’s replies, address one another by name, and finish with one shared written conclusion.

The vendors have to come from different companies — that is the whole point, not the settings. Models from one family were trained on similar data: they agree in the same places and get things wrong in the same places. A council of clones is one opinion said eight times, and it sounds more convincing than it deserves.

Specifications: wired together on n8n; the line-up is Gemini, Llama, Qwen, Gemma, Nemotron, GPT-OSS, Cohere and Poolside, all on free tiers. It lives on my own server and is used like an ordinary chat.

Nearby work

Back to Automation & AI agents