AgentDesk

Inspiration

Small businesses lose customers every day because they can't respond to inquiries fast enough. Hiring a support agent costs $2,000+/month — most small businesses can't afford that. I run a dental clinic (helloreeza.com) and experienced this firsthand. I built AgentDesk to solve my own problem first, then share it with other small businesses.

What it does

AgentDesk is an AI customer support widget that any business can embed with a single line of code. It uses Google Gemini 3 Flash Preview to handle customer conversations in real-time — answering questions about services, pricing, and hours. It books appointments by collecting date/time/service preferences, captures visitor leads (name, email), and auto-escalates frustrated customers to human agents.

How I built it

I built AgentDesk solo using Next.js 16 for the full-stack framework, Tailwind CSS 4 for styling, Framer Motion for animations, and Three.js for the 3D hero on the landing page. The AI is powered by Google Gemini 3 Flash Preview — I call the Gemini API directly from Next.js API routes, passing the business's knowledge base as context so the AI gives accurate, domain-specific responses. The database and auth use Supabase (free PostgreSQL), and payments are processed through Stripe. Deployed on Vercel.

Challenges I faced

Integrating Gemini AI to give business-specific answers (solved by injecting knowledge base data into the system prompt)
Making the Stripe checkout work with test keys (had a bug that blocked test mode)
Building a full SaaS product solo — landing page, dashboard, AI widget, billing — in 3 weeks

What I learned

The key to making AI useful for businesses is context. Generic AI gives generic answers. By feeding the AI your specific services, pricing, FAQs, and business hours, it becomes a knowledgeable support agent that actually helps customers. I also learned that shipping fast and iterating based on real usage matters more than perfect code.