Leap - AI Developer Agent That Deploys to Your AWS/GCP

TL;DR

TL;DR: Leap is an AI developer agent that builds and deploys real backend applications with proper infrastructure to your own AWS or GCP cloud — no more prototypes that can’t go to production.

What Is Leap?

Leap is a web-based AI coding agent that goes beyond typical prototype builders. It connects to your GitHub, works with pull request-style code revisions, generates architecture diagrams and API docs as it builds, and deploys to your own cloud infrastructure.

The core problem Leap solves: most AI app builders produce things that look impressive but fall apart when you need a real backend, proper test environments, and production-grade infrastructure.

How It Works

Leap runs on the open-source Encore.ts framework, which provides declarative infrastructure definitions — a cloud-agnostic CDK layer. This means both your application logic and infrastructure are defined in the same context, making AI code generation more reliable and less error-prone.

Step 1: Connect GitHub

You start by authorizing Leap to access a GitHub repository. Your code lives in a repo you control — no lock-in.

Step 2: Describe What You Want to Build

You describe the application in natural language. Leap generates code and presents it as a diff, similar to a pull request. You can review, request changes, and merge when satisfied.

Step 3: Architecture Diagrams and API Docs

As Leap builds, it generates architecture diagrams and API documentation for your app. This gives you full visibility into what was created.

Step 4: Isolated Preview Environment

Before deploying, your app runs in an isolated preview environment. Test changes without affecting production.

Step 5: Deploy to Your Cloud

When ready, deploy to your own AWS or GCP account using Encore’s open tooling. No proprietary deployment platform required — your cloud, your rules.

Key Features

PR-style code revisions — Code changes are presented as diffs you can review before accepting. Not a black box; you see exactly what changes.

Isolated preview environments — Each change gets its own preview environment. Test thoroughly before promoting to production.

Architecture diagrams — Generated automatically as the app is built. No more reverse-engineering your own codebase to understand its structure.

Full infrastructure ownership — Encore.ts handles declarative infrastructure (databases, queues, services) as TypeScript code. Infrastructure is version-controlled alongside your application.

Open-source framework — The underlying Encore.ts framework is open source. Run locally with the Encore CLI and build Docker containers without vendor lock-in.

Technical Details

Leap uses Claude 4 Sonnet as its reasoning engine. The choice of Encore.ts as the framework was deliberate — declarative infrastructure definitions make it naturally suited for AI code generation. When both application logic and infrastructure are defined programmatically in the same language context, the model has better visibility into what it’s generating and can produce more coherent results.

The biggest current limitation: larger codebases still challenge the model’s context window. The team acknowledges this and has designed Leap so you can switch to your local IDE at any time and continue working there. For local development, you only need the open Encore CLI.

Practical Evaluation Checklist

• GitHub integration — Does it actually connect and create proper repos? Yes, OAuth GitHub integration is the first step.
• Code quality — PR-style diffs mean you can review before accepting. Architecture diagrams give visibility.
• Infrastructure — Declarative TypeScript infrastructure via Encore.ts. No hidden infrastructure or proprietary abstractions.
• Deployment — Deploys to your own AWS/GCP. Open-source Encore CLI for local runs.
• Preview environments — Isolated per-change previews. Good for testing before production.
• Context limitations — Works best for new projects or isolated services. Large existing codebases are still challenging.

Security Notes

Code lives in your GitHub repository — you retain full ownership and access. Infrastructure is defined in your own cloud accounts. No proprietary runtime that could lock you in.

FAQ

Q: Does Leap work with existing projects or only new ones? A: Currently best for starting new projects or building isolated services within larger systems. Large existing codebases still struggle with context length.

Q: What cloud providers are supported? A: AWS and GCP are the primary targets. Encore.ts is cloud-agnostic, but the integration currently focuses on these two.

Q: Is the code open source? A: Yes — the Encore.ts framework is open source (MIT license). You can run your app locally with the Encore CLI and build Docker containers without Leap.

Q: How does it compare to other AI coding agents? A: Most AI app builders are prototype-focused. Leap differentiates by producing deployable backend infrastructure with proper isolation, version control integration, and architecture documentation.

Conclusion

Leap tackles a real frustration in the AI coding tool space: the gap between “impressive demo” and “production-grade backend.” By building on Encore.ts’s declarative infrastructure model, it can generate both application logic and infrastructure definitions in the same context — reducing the error rate that plagues other LLM code generation attempts.

The PR-style review flow, architecture diagrams, and isolated preview environments make it more developer-friendly than typical black-box AI builders. And deploying to your own cloud (rather than a proprietary runtime) means you actually own what you build.

It is early, and the context window limitation for larger codebases is a real constraint — but the team has designed the workflow so you can seamlessly switch to your IDE when needed. If you have been burned by AI coding tools that look great but cannot actually ship to production, Leap is worth a try.

Links:

• Site: https://leap.new
• Launch: https://news.ycombinator.com/item?id=44137177