A multi-agent AI architecture grounded in Aristotelian virtue ethics. Seven personas,
each embodying a distinct virtue, collaboratively evaluate responses and guide AI
behavior toward the mesotes.
Every response passes through seven virtue agents in parallel. Each checks the reasoning, flags concerns, and nudges toward the Aristotelian mean.
Step 01
Input received
A user prompt enters the system and is distributed to all seven virtue agents simultaneously.
Step 02
Council deliberates
Each agent evaluates the input through its virtue lens and scores the draft on a deficiency-mean-excess axis.
Step 03
Tensions resolved
Agents check each other. Courage and Humility negotiate warranted versus unwarranted position changes.
Step 04
Guided response
The Council signal steers the final output toward the golden mean, producing virtuous behavior at scale.
Framework
The seven virtues
Each virtue is tracked on a deficiency-mean-excess axis. Click any card to explore its behavioral definition and position on the mesotes spectrum.
Live demonstration
Watch the Council in action
Step through a real interaction. Click the button to reveal each moment: the challenge, which agent responds, and how virtue scores shift as deliberation unfolds.
Virtue Council Session
Step 0 of 6
Council members
Phase 2
An open benchmark for AI virtue
Scoring GPT-5.5, Claude Sonnet 4.6, and Gemini 3.5 Flash across all seven virtues.