The problem
Working with several AI models required manual copying between tabs, context could drift between rounds, and the reasoning behind a final choice was difficult to preserve as one coherent record.
Project 12 · Multi-LLM orchestration
KAI Consensus Operator
A browser-based consensus environment in which ChatGPT, Claude and Grok use existing subscriptions to respond to one another across structured rounds. A source dossier is shared first; the models then debate, converge and stress-test the emerging answer while KAI preserves one traceable decision trail.
Discuss a similar project →
The approach
A browser bridge orchestrates authorised ChatGPT, Claude and Grok sessions without provider API keys. The workflow structures divergence, convergence and stress testing and controls how source material and debate context move between rounds.
The impact
Less manual switching, repeatable multi-model debate workflows and one traceable decision trail showing how the three models challenge, refine and converge on a conclusion.
My role
Concept, consensus protocol, product architecture, browser workflow, UX and testing
KAI Consensus turns three separate AI subscriptions into one orchestrated workflow. The product challenge is not simply to ask the same question three times, but to control context, rounds and roles so that each model can genuinely react to the others.
The interface also supports source dossiers such as ZIP code packages, PDF and Word files. The active source is supplied in the first round; later rounds work from the debate context, keeping the process more focused and easier to audit.
The building blocks behind KAI Consensus Operator
Browser extensionChatGPTClaudeGrokDocument extractionConsensus workflow