Story

I posted this yesterday and kept building. Original post got some good feedback about wanting more domain-specific personas beyond the generic analyst/contrarian archetypes. So I added them. 35 personas now, organized into actual professional structures:

Law Firm — Litigator + Corporate Counsel + Compliance Officer + Junior Associate, synthesized by a Senior Partner Hospital Team — GP + Specialist + Pharmacist + Medical Ethicist, synthesized by Chief of Medicine Editorial Team — Reporter + Editor + Legal Reviewer + SEO lead, synthesized by Editor-in-Chief Corporate — CFO + CTO + CMO + Legal, synthesized by CEO Startup — Founder + Engineer + Designer + Growth Lead, synthesized by Investor Consulting — Strategy + Operations + Finance + Risk, synthesized by Senior Partner

Each persona has a role-specific system prompt tuned to how that function actually thinks — the CFO talks in EBITDA and burn rate, the Junior Associate flags the clause the partners missed. Also shipped in v2:

Temperature slider per run (Precise → Balanced → Creative) Follow-up questions — council carries the full prior verdict as context Abort mid-run with partial result preservation Export MD / PDF Import/export council config as JSON Webhook after every completed session (works with Zapier, n8n, Make)

Still zero backend. Still runs entirely in the browser. API keys never leave your machine.

GitHub: https://github.com/prijak/Ai-council

Livelink: https://council.gameinghub.com/

Genuinely curious: has anyone found multi-model deliberation actually useful for a specific domain, or does it mostly just produce longer wrong answers?