llm-council

To run a multi-model LLM deliberation, you invoke a four-model Claude council workflow in-process. Four models independently answer your architecture prompt, anonymously peer-review each other, and an Opus president synthesizes a final verdict preserving dissent.

Anonymized peer-review in multi-model LLM workflows is a blind labeling process where multiple Claude models evaluate each other's independent responses without knowing the source. This surfaces true differences and biases before final synthesis.

No, you do not need external API keys or network calls to run a Claude deliberation council. The workflow requires only in-process Workflow agents for four Claude models to perform the independent responses and synthesis.

Use a four-model council instead of a single LLM prompt for high-stakes decisions, complex verification tasks, and risk-sensitive architecture choices. The deliberation process captures diverse perspectives and preserves minority views that a single prompt would miss.

Yes, the deliberation council workflow works directly within a Claude Code session. You invoke the workflow in-process to execute the four-model deliberation and receive the synthesized verdict without external dependencies.

The best way to preserve dissent in LLM-generated architecture reviews is using a deterministic in-process synthesis by an Opus president model. It synthesizes the final verdict from anonymized peer-reviews while intentionally preserving minority views.