Magic – Instant Secure Full‑Stack Apps & AI Agents

What it is

Magic is an open‑source platform that lets you turn a plain‑English description into a complete, self‑hosted full‑stack application. It automatically creates:

• a relational database schema (or connects to an existing one),
• a secured REST API with role‑based access control (RBAC),
• a React‑based dashboard for managing the app, and
• optional AI‑driven agents that can call the generated endpoints.

All components run on your own hardware – there is no SaaS lock‑in. The core runtime is Hyperlambda, a compiled .NET‑based language that can be generated by an LLM and executed inside a sandbox.

Key features (from the README)

Feature	What you get
One‑click start	`curl …
Natural‑language generation	Describe a database, an endpoint, or UI in plain English; the built‑in generator writes Hyperlambda, SQL, HTML, or CSS for you.
Secure, self‑hosted backend	Compiled .NET runtime, sandboxed Hyperlambda, per‑function whitelist, RBAC, JWT auth, OIDC sign‑in (Google, GitHub, Azure AD, etc.).
MCP server	Every endpoint is exposed as a Model Communication Protocol tool, so Claude, Cursor, Codex, Qoder, etc., can invoke them directly.
AI‑agent friendly	Agents can query/update your own databases, call your own APIs, and run background jobs – all behind your own access control.
Import third‑party APIs	Paste an OpenAPI/Swagger URL; Magic generates typed, documented CRUD‑style wrappers with automatic validation.
Sandbox API	Generates an endpoint that executes caller‑supplied Hyperlambda (or English‑generated code) inside a role‑based whitelist – safe “run‑code‑as‑a‑service”.
Web Designer	In‑browser WYSIWYG editor that edits the actual HTML file on disk; 60+ pre‑built blocks, live preview, and AI‑assisted editing.
Chat Ops	Global “Ctrl+.” shortcut opens a chat window that can ask the server to generate code, run functions, or retrieve data – the model’s actions are executed server‑side and shown with real‑time status pills.
Rewind debugger	Record an execution, then step forward/back through each statement with full variable values displayed.
Performance	Claims 20× faster than FastAPI/Flask, 50× faster than LangChain, and up to 1 000× faster than workflow tools because Hyperlambda compiles to native .NET code.
Deployment options	Docker Compose (quick start), DigitalOcean droplet (single‑node), Kubernetes (cluster), or run from source (.NET 10 + Node).
AI model flexibility	Uses OpenAI by default but also supports Ollama, HuggingFace, and a local BM25 retrieval fallback that needs no embeddings.
Extensible language	Hyperlambda can call C# functions, run Python scripts, Bash commands, or use PuppeteerSharp for headless‑browser automation.

How it works (high‑level flow)

1. Start the stack – Docker Compose launches a backend container (magic-backend) and a frontend container (magic-frontend).
2. Log in – Default admin user root/root (or configure OIDC). A JWT is issued for each session.
3. Create a project – Use the Endpoint Generator or API Wizard in the dashboard. Point it at a database (MySQL, PostgreSQL, SQL Server, MariaDB) or an OpenAPI spec.
4. Describe what you need – Type a natural‑language prompt in the “Machine Creates the Code” bar. The LLM generates an AST (Abstract Syntax Tree) for Hyperlambda; the system validates that every called function exists before saving.
5. Run / debug – Execute the generated file directly from the IDE, or click Debug to record a step‑by‑step trace.
6. Expose as MCP tools – The MCP plugin automatically registers each HTTP endpoint as a tool that any compatible LLM can call, cutting token usage by ~80 %.
7. Deploy – Since everything runs on your machine, you simply expose ports or put the containers behind a reverse proxy (Caddy is pre‑configured for HTTPS on DigitalOcean). No additional build or publish step is required.

Who might use it

• Solo developers or small teams who want to prototype a CRUD app quickly without writing boilerplate code.
• Enterprises that need to keep data on‑premises but still want AI‑assisted development and internal agents.
• AI‑ops teams building custom agents that must interact with legacy databases or internal services securely.
• Educators / hackathon participants looking for a demo‑ready platform that showcases AI‑generated code and secure sandbox execution.

Limitations / considerations (as stated in the README)

• The LLM can still produce logically incorrect code – only function existence is guaranteed.
• Security relies on correct RBAC configuration and the whitelist; mis‑configuration could expose dangerous functions.
• Performance claims are based on the authors’ benchmarks; real‑world results may vary.
• The platform is relatively new; community support and ecosystem (plugins, extensions) are still growing.

Quick start recap

curl -fsSL https://hyperlambda.dev/docker-compose.yaml | docker compose -f - up
# then open http://localhost:5555, log in with root/root

From the dashboard you can immediately point the API Wizard at a sample chinook database and watch 50+ secured endpoints appear.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.