polterguy/magic

What it solves

Magic allows users to build full-stack applications and AI agents using plain English, eliminating the need for traditional build and deploy cycles. It solves the problem of vendor lock-in and high token costs associated with proprietary AI app builders by providing a self-hostable, open-source alternative that runs on the user's own hardware.

How it works

The platform uses a proprietary LLM to generate Hyperlambda, a language that produces an Abstract Syntax Tree (AST) rather than raw text, ensuring that only existing functions are invoked and preventing hallucinations. Hyperlambda runs in a secure, sandboxed environment with a role-based access control (RBAC) system that whitelists individual functions. The system includes a dashboard with a Hyper IDE, SQL Studio, and an Endpoint Generator to turn databases or OpenAPI specs into secured REST APIs. It also features an MCP server that exposes these endpoints as tools for AI agents like Claude or Cursor.

Who it’s for

It is designed for software developers and business owners who want to rapidly prototype and deploy CRM systems, admin panels, internal tools, and database-driven AI agents without sacrificing control over their data or infrastructure.

Highlights

• Zero-Deploy Workflow: Save and run code immediately without a build or deploy step.
• MCP Integration: Built-in MCP server that turns API endpoints into tools for AI agents, reducing token consumption by approximately 80%.
• Secure Execution: Sandboxed runtime that prevents unauthorized file system access and allows function-level whitelisting.
• RAG-powered Chatbots: Includes a crawler and three retrieval modes (semantic, keyword, and hybrid) to create grounded AI expert systems.
• High Performance: Compiled .NET runtime that is significantly faster than interpreted workflow tools or frameworks like FastAPI and LangChain.
• Rewind Debugging: Ability to step through recorded executions to see exactly how program state evolved.