dsh-ai-council

Dropping several models into the same prompt for a quick vote sounds efficient, but it often breeds anchoring and manufactured consensus where no one genuinely challenges from a security, performance, or commercial angle. dsh-ai-council turns DeepSeek Harness into a temporary corporate decision board: every seat carries its own professional mandate and authority, and the model is simply the person in the chair.

The protocol works in three phases. In Round 1 every member drafts an independent structured position with no visibility into peers, eliminating premature convergence at the source. Later rounds expose the prior focus and unresolved disputes; members must address objections head-on and are explicitly told not to agree just to manufacture consensus. The Chair then checks weighted approval thresholds and authoritative role blockers before declaring a real consensus. If the bar is met, deliberation stops early; at the maximum round the Chair either adjudicates with evidence or defers explicitly. A failed model route replaces the model, never the role.

It is built for teams that need structured, auditable technical decisions: architecture reviews, production-readiness gates, product launches, security and compliance sign-offs. Twenty corporate role presets and eight decision templates ship out of the box, all editable in the Settings UI, so every multi-model collaboration reads less like a ballot box and more like a boardroom.