I built TaskWand (https://taskwand.io) to make creating n8n workflows faster and less error-prone.

The Problem I love n8n, but building complex workflows from scratch takes time. While standard LLMs can suggest logic, they often "hallucinate" nodes that don't exist, get parameter names wrong, or output JSON structures that n8n can't actually import.

The Solution I built a specialized RAG (Retrieval-Augmented Generation) system. I indexed two thousands of working, verified n8n workflows. When you describe a task, the system retrieves relevant, valid workflow snippets to ground the LLM's response. This significantly reduces hallucinations and ensures the output is import-ready.

Key Features:

Visual Preview: You don't just get JSON; I built a UI that renders the n8n workflow preview directly in the browser so you can verify the logic before exporting.

Prompt Refiner ("Improve"): Writing detailed technical specs is hard. This button takes a vague idea (e.g., "sync hubspot to slack") and rewrites it into a highly detailed, technical prompt optimized for the generator.

Interactive Context ("Ask"): A Q&A copilot to answer questions about nodes and logic or troubleshoot concepts before generation.

The Tech Stack I decided to use a bleeding-edge stack for this:

Frontend: Next.js and Tailwind CSS.

AI/RAG: OpenRouter API (GPT models) and Qdrant (Vector DB).

Backend: Supabase (Auth & DB), Next.js Serverless Functions.

UI Components: react-markdown, react-syntax-highlighter, and the official n8n component for the visualization.

I’d love to hear your feedback on the generation quality and the UI experience.