Duo Chat AI — Open-Source Dual-Agent Dialogue

Inspiration

Most AI apps speak in one voice. Duo Chat AI lets two agents debate, critique, and co-reason—useful for students, researchers, and anyone who learns best from dialogue. We open-sourced it to make multi-agent reasoning accessible in classrooms, labs, and communities.

What it does

• Dual-agent conversations: agents debate, question, and fact-check each other
• Cross-model: run 20B vs 120B (gpt-oss) or mix backends
• Local or online: privacy-friendly, offline-capable
• Modern UX: personas, chat export, responsive UI

How we built it

• Frontend: React 18, TypeScript, Vite, Tailwind, Radix UI
• Inference: gpt-oss-20B (default), optional gpt-oss-120B via vLLM, Ollama, LM Studio
• Pipeline: dual-agent orchestration, JSON streaming, function calling, multi-model routing

Challenges

• Keeping debates balanced (avoiding one agent speaking “for” the other)
• Smooth multi-backend integration (vLLM/Ollama/LM Studio)
• CORS and cross-platform setup
• Throughput vs depth (20B speed vs 120B depth)

Accomplishments

• Working dual-AI debate system on a hackathon timeline
• Accessible, mobile-friendly UI
• Proof that open gpt-oss is strong at structured reasoning
• Released under Apache-2.0

What we learned

• Multi-agent dialogue boosts learning and critique
• Local-first setups matter for privacy/offline communities
• Open weights & OSS speed up real-world adoption

What’s next

• Fine-tuned personas (Teacher, Debate Coach, Research Mentor, Storyteller)
• Lighter models for edge devices
• Collab tools: shared sessions, annotation, classroom mode
• Accessibility: voice I/O, translation layers
• Community roadmap with educators/researchers/devs

“Duo Chat AI isn’t a chatbot—it’s a stage where two lines of reasoning meet.”

Search Keywords

Duo Chat AI, dual agent AI, multi-agent AI, AI debate, Socratic AI, cross-model conversation, model vs model, LLM vs LLM, open-source AI chat, Apache-2.0 AI, gpt-oss 20B, gpt-oss 120B, structured reasoning, collaborative reasoning, adversarial collaboration, critique generation, debate training, education technology AI, classroom AI tools, research assistant AI, peer review AI, tutoring AI, study companion AI, prompt engineering, persona management, multi-persona chat, JSON streaming, function calling, tool use, multi-model routing, vLLM, Ollama, LM Studio, local inference, offline AI, privacy-preserving AI, on-device AI, edge AI models, React AI app, TypeScript AI UI, Vite, Tailwind CSS, Radix UI, responsive AI UI, chat export, chat history export, mobile-friendly AI, desktop web app, open weights models, community AI project, academic use AI, hackathon project AI, debate coach AI, research mentor AI, creative writer AI, classroom mode, annotation tools, teacher tools AI, learning science AI, educational debate, AI sandbox, reproducible AI, OSS ecosystem, plug-and-play LLMs, backend agnostic AI, reasoning-first models, evaluation harness, latency vs quality tradeoff, CORS setup, cross-platform dev, multi-device sync, persona editor, dark mode light mode, modern UI, transparent AI, democratizing AI, open research, collaborative learning, peer learning, dialogic learning, explainable reasoning, AI for schools, AI for universities, AI for labs, AI for communities

Built With

• gpt-oss
• lmstudio
• ollama
• radixui
• react
• tailwindcss
• typescript
• vite
• vllm