Welcome to Crucible
Crucible is an AI decision-intelligence platform that convenes multiple AI Agents to debate strategic questions and produce defensible Decision Briefs. This guide will help you get started.
What is Crucible?
Crucible forces multiple AI Agents to challenge each other's claims, burn away errors, and converge on defensible conclusions. Instead of accepting the first answer an AI gives, we orchestrate structured debate across up to 12 expert Agents.
Each session produces two key artifacts:
- Decision Brief - An executive-ready recommendation with citations, risks, and next actions
- Meeting Minutes - A complete audit trail of claims, challenges, and evidence
Core Concepts
Agents
AI Agents (formerly called Knights) are specialized expert personas with distinct roles, prompts, and preferred models. You can use official Agents curated by Crucible, browse community Agents from the Marketplace, or create your own.
Sessions
A session is a structured debate where Agents research, present claims, challenge each other, and converge on a recommendation. Sessions typically complete in 45-60 minutes and can host up to 12 Agents.
Artifacts
Every session produces audit-ready artifacts including the Decision Brief (executive summary) and optional Meeting Minutes (full transcript). Artifacts include citations, areas of dissent, and confidence scores.
Quick Start Guide
Frame Your Question
Start with a strategic decision at stake. Share context, constraints, and supporting documents so every Agent joins with full context.
Tip: Avoid yes/no prompts. Ask for trade-offs, risks, and desired outcomes so the debate uncovers real options.
Select Agents
Crucible suggests Agents based on your question. You can swap in community or private Agents when you need specialist judgment.
Each Agent comes with a role, prompt, and preferred model. Seat up to twelve per session for a balanced debate.
Run Structured Debate
Three rounds keep sessions tight: Research, Claims, Challenges, and Convergence. The moderator enforces turn-taking and token discipline.
Use guardrails (retrieval, policy filters, dissent quotas) to align with your risk posture.
Review Artifacts
Every session produces an audit-ready Decision Brief plus optional minutes. You get the converged recommendation, dissent, evidence links, and next steps.
Export artifacts to PDF or sync with your workspace via API (coming soon).
Next Steps
Helpful Resources
Need deeper docs? Email us at hello@roundtablelabs.ai with your request.