IdeaForge

Inspiration

Most ideas fail because they aren't pressure-tested. We built IdeaForge to automate structured evaluation using AI agents that debate ideas from multiple perspectives. Like a virtual team meeting, but faster.

What it does

Enter a business idea. IdeaForge orchestrates 4 AI agents (Role Selector, Adversary, Ally, Synthesizer) representing different roles. They debate across two rounds, identifying risks, proposing mitigations, and refining the plan. Output includes a refined plan, risk analysis, MVP scope, KPIs, and a go/no-go verdict. All in minutes.

How we built it

Tech Stack: Frontend: Next.js 14, TypeScript, Tailwind CSS Backend: Next.js API routes, Toolhouse AI agents Architecture: 5-step orchestration (Role Selection → Debate Round 1 → Debate Round 2 → Synthesis → Export)

Challenges we ran into

Toolhouse API integration: Agents require GUIDs and all variables must be strings Template variable handling: Proper vars object formatting for agents with templates JSON parsing: Handling markdown-wrapped JSON responses from agents Error handling: Graceful fallbacks when individual agents fail

Accomplishments that we're proud of

Getting the multi-agent orchestration working after debugging Toolhouse API integration
Seeing the debate process surface risks we hadn't considered
Building something that actually helps evaluate ideas, not just demo tech
Creating a clean UI that makes the complex process feel simple ## What we learned

Toolhouse agent configuration and template variable requirements AI orchestration patterns for coordinating multiple agents Structured debate methodology (adversary → ally) surfaces better insights Type safety with dynamic AI responses requires robust parsing

What's next for IdeaForge

Agent customization and role configuration Historical comparisons and idea tracking Team collaboration features Integration with project management tools (Jira, Notion) Industry-specific evaluation templates Real-time collaborative workshops