Inspiration

PromptTester wasn’t born from a roadmap — it started as a curiosity. I opened up Bolt.new to play with ideas, and five days later, I had a working IDE. I had no coding background, no intention to go all the way. But once I started designing how prompt engineering should feel — modular, testable, visual — I couldn’t stop until it was real.

As AI systems evolve rapidly, prompts are becoming the new software. Yet, tools for serious prompt work remain either too simple or too technical. I wanted to change that — to create the VS Code for prompt engineers.

What It Does

PromptTester helps prompt engineers build smarter and faster:

• Modular prompt building with reusable text/data blocks
  
• Variant testing with dynamic datasets
  
• Visual interface to compare outputs across models and runs
  
• Secure API key handling using encrypted Supabase Edge Functions
  
• Full support for both text and image models (OpenAI, Replicate, Anthropic, Google)
  
• Real-time previewing and fast iteration for creative and technical workflows

How It Was Built

• Frontend: Vite + React + Tailwind + Shadcn/UI for sleek performance
  
• State: Zustand for lightweight, reactive state logic
  
• Drag-and-Drop: Powered by Dnd-kit for intuitive block handling
  
• Backend: Supabase for auth, database, and secure server-side Edge Functions
  
• Model Integration: Flexible OpenAPI-based routing to connect any LLM, including image models
  
• Security: API keys never touch the client; all traffic goes through encrypted, authenticated Edge Functions

And all of this was built through prompts — I didn’t write a single line of code from scratch.

Challenges We Overcame

• Making sure no API keys ever touch the client — encryption and Row Level Security in Supabase was crucial
  
• Integrating with inconsistent API specs from multiple model providers
  
• Designing a prompt builder UI that felt both powerful and intuitive
  
• Handling real-time completion comparisons, asynchronous workflows, and responsive updates
  
• Adapting the system for image models, which often break the mold in terms of payload and pricing

Accomplishments We're Proud Of

• API keys are safe. Full stop — encrypted at rest, never exposed
  
• Prompt blocks are reusable, testable, and visual — a productivity booster for anyone working with LLMs
  
• Image support looks incredible and works across providers
  
• Users can visually compare prompt results side-by-side in seconds — it’s fast and satisfying
  
• All of this was built in days, by someone who just got obsessed with getting it right

What We Learned

• API integration is never “just plug it in.” Every LLM is its own beast
  
• Security isn’t optional — users deserve trust by default
  
• State management matters more than you think when UIs get dynamic
  
• Bolt.new + AI is ridiculously powerful when you pair it with just enough stubbornness and taste
  
• The best products often start as accidents

What’s Next for PromptTester

• Prompt Version Control – track changes, revert, collaborate
  
• Advanced Evaluation – A/B testing, auto-metrics, custom scoring
  
• Team Features – shared projects, comments, role permissions
  
• Production Deployment – export prompts to pipelines, track usage
  
• More Models – expand into coding agents, niche tools, and domain-specific LLMs

Built With

• anthropic-api
• authentication
• dnd-kit
• edge-functions)-deployment:-netlify-llm-integrations:-openai-api
• google-gemini-api
• lucide
• mistral-ai-api
• netlify
• openai
• react
• react-hook-form
• replicate
• shadcn/ui
• sql-(postgresql)-frontend-frameworks/libraries:-react
• supabase
• tailwind-css
• typescript
• vite
• zod-icons:-lucide-react-backend/database:-supabase-(postgresql
• zustand