Petamind

About

Petamind is an advanced MCP server designed for Claude Code, implementing a sophisticated multi-candidate agentic coding loop. It functions by generating various patches based on a reasoner plan, subjecting them to deterministic gates and mandatory vision scoring, and ultimately selecting the best solution. Beyond its core agentic coding capabilities, Petamind also includes a production-grade synthetic dataset generator, ideal for creating UI/UX design tasks for landing pages, directories, and dashboards using modern web technologies like Next.js, TypeScript, and Tailwind.

Key Features

• 1 GitHub stars
• Multi-model pipeline supporting Vertex AI (DeepSeek, Kimi, MiniMax) and OpenRouter (Devstral, vision models)
• Quality gating ensuring only high-scoring, build-successful candidates proceed to training data
• Resumable operations with SQLite caching for model responses and task state persistence
• Generates production-grade synthetic UI/UX datasets for various web page types
• Strict separation of chain-of-thought/thinking from structured specifications and final code

Use Cases

• Creating high-quality synthetic UI/UX datasets for training AI models in design tasks
• Automating multi-candidate patch generation and validation within agentic coding workflows
• Iteratively refining code solutions using vision scoring and deterministic build gates