Hyperlambda
Deterministic Output
Hallucination-Resistant Execution
The model can propose structures, but only valid, whitelisted AST nodes can execute.
Natural language in, deterministic AST out
Instant Full-Stack Apps and AI Agents. Secured by the Runtime.
Describe what you want in English. Hyperlambda compiles it to a strict AST that executes only what the whitelist permits — secured APIs over your databases, frontends to match, complete in seconds for cents.
# When it's up: log in at http://localhost:5555 with root/root# Change that password before the internet does.$ curl -fsSL https://hyperlambda.dev/docker-compose.yaml | docker compose -f - upC# Active Events Sandbox
Structurally Constrained by Design
Hyperlambda is not free-form generated source code. It is a constrained AST validated before execution.
Self-Extending
Evolving AI Agents
Create AI agents that evolve over time with new tools, workflows, APIs, and backend capabilities generated on demand.
How it works
From English to safe execution
Run Magic Cloud locally, describe what you want in English, and Hyperlambda compiles it into strict ASTs for deterministic execution.
Step 1
Describe the task
Write what you want in natural language
Step 2
Compile to Hyperlambda
The compiler turns your request into Hyperlambda AST
Step 3
Execute safely
Magic Cloud runs the result inside a constrained C# runtime
Hyperlambda runs in a sandbox that can restrict the execution layer at individual function level — which makes AI-generated code safe to execute by construction: a hallucinated capability fails before it runs.
Concrete example
From English to Safe Backend Execution
“Create an endpoint that returns unpaid invoices for a customer.”
Generated Hyperlambda AST
.argumentscustomer_id:longdata.connect:billingdata.readtable:invoiceswhereandcustomer_id.eq:x:@.arguments/*/customer_idstatus.eq:unpaidreturn-nodes:x:@data.read/*Runtime result
[{"id": 42,
"customer_id": 7,
"invoice_number": "INV-10042",
"amount": 1299.00,
"due_date": "2026-05-15",
"status": "unpaid"},
{"id": 57,
"customer_id": 7,
"invoice_number": "INV-10057",
"amount": 249.50,
"due_date": "2026-06-01",
"status": "unpaid"}]Safety
- The model proposed the AST.
- Magic Cloud validated the AST before execution.
- Only whitelisted nodes were available.
- Unauthorized nodes would fail before execution.
- The runtime, not the AI model, is the security boundary.
Why this matters
- Traditional AI code generation produces free-form source code.
- Hyperlambda produces constrained structures.
- Constrained structures can be validated before execution.
- This makes AI-generated backend automation more predictable and easier to secure.
Full-stack, not just endpoints
From one prompt to a working application
This carbon calculator is a complete application on a Magic cloudlet — SQLite database, public and role-gated API endpoints, and the frontend itself, all served from the same system. An AI agent built it autonomously during a live client meeting: database, endpoints, SPA — deployed and smoke-tested before the call ended.
- One system, one deployment — database, secured API, and frontend together.
- Generated by an agent from a natural-language description.
- Custom branding and design — a real product, not a template.
Click the screenshot to open it full size.
The dashboard
Everything your cloudlet can do, in one place
Every Magic Cloud install ships with a dashboard for building and running your backend.
- Hyper IDE — edit and execute any file, with Hyperlambda autocomplete.
- SQL Studio & Generator — design SQLite, MySQL, PostgreSQL and SQL Server, and turn tables into secured CRUD APIs.
- MCP & Machine Learning — expose endpoints to any MCP harness, and train embeddable chatbots on your own content.
Click the animation to open it full size.
Take the full dashboard tour
How we compare
Honest comparisons
Feature matrices written the uncomfortable way — including the rows we lose.
See it in action
From prompt to Full Stack in 7 minutes
In this video I demonstrate how to create and deploy into production a full stack app in 7 minutes, 100% on requirements, and exactly what the client wanted.
Everything in the video ships with every install — MIT-licensed, self-hosted, and running on your own machine minutes after the Docker command at the top of this page.
What developers say
Developers on Magic and Hyperlambda
“Kudos for the incredible work: it's the first real IDE dedicated to AI.”
Samuele ContardiSoftware Architect“The truth is that I like Firebase, but your system is made much simpler.”
Alex ErdeiFrontend Developer“It's so amazing and revolutionary, yet easy to use and elegant.”
Justin LietzSoftware EngineerFrequently asked questions
What is Hyperlambda?
Hyperlambda is a programming language whose code is a tree structure rather than free-form text — effectively an executable AST. That structure is what lets natural-language requests compile into something deterministic and inspectable before it runs, which is why it suits AI-generated backends.
What can I actually build with it?
Complete full-stack applications — secured CRUD APIs over your databases, frontends and SPAs served from the same cloudlet, backend workflows, scheduled tasks, AI agents and tools, and embeddable chatbots trained on your own content. All generated from short natural-language prompts rather than hand-written.
Can it build complete applications, including the frontend?
Yes. A Magic cloudlet serves static sites and single-page applications alongside the APIs they consume, so a working prompt-to-app flow is: design the database, generate the secured endpoints, and deploy the frontend — one system, one deployment. Users, roles, and full OIDC and OAuth support are built into the platform, so the applications you ship come with real authentication rather than something bolted on per project.
Do I need to know Hyperlambda to use it?
No. You describe what you want in plain language and the generator compiles it to Hyperlambda for you. Knowing the language helps you read and adjust the result, but it is not required to build working backends.
What makes AI-generated code safe to run?
The security boundary is the runtime, not the model. Generated Hyperlambda compiles to an AST whose nodes can only bind to explicitly whitelisted capabilities in the current context; anything outside that fails before it executes. The model proposes, the runtime decides.
Does it work with any LLM or AI coding agent?
Yes. The runtime's safety model is independent of which model generated the code. Magic integrates OpenAI, Ollama and HuggingFace for its own AI features, and with the MCP plugin installed your cloudlet exposes an MCP server that any MCP-capable harness — including Claude, Codex, Cursor and Qoder — can connect to, using your endpoints as tools.
What databases does it support?
SQLite is built in, and you can connect external MySQL, PostgreSQL and Microsoft SQL Server. Once connected, the platform treats an external database like a local one for querying, design and code generation.
Which sign-in (SSO) providers does it support?
Magic ships single sign-on out of the box for Google, GitHub, LinkedIn, Microsoft Entra ID, Okta, Auth0, Keycloak and Slack. Each provider is turned on from the dashboard with a client ID, external identities are scoped per provider and mapped onto Magic's own role-based access control, and a provider is a single Hyperlambda file — so adding another is dropping a file into a folder.
How is Magic different from Supabase, n8n, or PocketBase?
Short version: Supabase is a hosted-first BaaS where self-hosting is the second-class path; n8n is a workflow orchestrator, not a backend; PocketBase shares Magic’s MIT single-binary philosophy but bets on humans writing Go rather than agents generating code. We publish honest feature matrices for each — including the rows Magic loses: Supabase, n8n, PocketBase.
How fast can it generate a working backend?
A single endpoint or tool typically generates in seconds. Because the output is validated structure rather than loose source code, what comes back is immediately runnable instead of needing a debug-and-fix loop.
Is Hyperlambda open source and free to use?
Yes. Magic Cloud, which runs Hyperlambda, is MIT-licensed and open source on GitHub. You can run the whole platform on your own hardware with no license fee.
Do I have to use a cloud, or can I self-host it?
You can self-host it entirely. Magic Cloud runs on your own server — a single Docker command brings up the full stack — and nothing leaves your machine unless an integration you configure sends it somewhere.
Is this production-ready, and what happens if the project dies?
Magic Cloud has run in production for years, and the whole platform is MIT-licensed with nothing held back — so the exit scenario is that you keep running the code you already have, fork it if you wish, and owe nobody anything. There is no license to renew, no hosted dependency to lose, and no vendor whose death takes your backend with it.
Who builds this
Thomas Hansen — creator of Hyperlambda
I wrote my first line of code at eight years old, and 44 years later I am still at it. Hyperlambda is the part I care most about — a language I created because I believe the answer to unsafe AI code generation is not better prompts, but a runtime that cannot be talked into doing the wrong thing.
Magic Cloud has been running in production for years, and every line of it is MIT-licensed and public. That is the deal: you can read everything, run everything on your own hardware, and the code stays yours no matter what happens to me. The bounty below is not marketing — I offer it because I know how the sandbox is built.
I dare you to break the sandbox
Escape the Hyperlambda execution sandbox from the playground — make generated code do something the whitelist forbids — and I will pay you $100. Nobody has.
Or just build something
One Docker command. Five minutes. Your hardware.
# When it's up: log in at http://localhost:5555 with root/root$ curl -fsSL https://hyperlambda.dev/docker-compose.yaml | docker compose -f - upPrefer it hosted? AINIRO runs cloudlets — same code, nothing gated.