AutoBE - AI backend builder for prototype to production

Describe your backend requirements in natural language through AutoBE's chat interface.

AutoBE will analyze your requirements and build the backend application for you. The generated backend application is designed to be 100% buildable by AI-friendly compilers and ensures stability through powerful e2e test functions.

With such AutoBE, build your first backend application quickly, then maintain and extend it with AI code assistants like Claude Code for enhanced productivity and stability.

AutoBE will generate complete specifications, detailed database and API documentation, comprehensive test coverage for stability, and clean implementation logic that serves as a learning foundation for juniors while significantly improving senior developer productivity.

Check out these complete backend application examples generated by AutoBE:

https://github.com/user-attachments/assets/b995dd2a-23bd-43c9-96cb-96d5c805f19f

To Do List: todo
Reddit Community: reddit
E-Commerce: shopping
ERP System: erp
- Requirements Analysis: Report
- Database Design: Entity Relationship Diagram / Prisma Schema
- API Design: API Controllers / DTO Structures
- E2E Test Functions: test/features/api
- API Implementations: src/providers

Getting Started

git clone https://github.com/wrtnlabs/autobe --depth=1
cd autobe
pnpm install
pnpm run playground

To use AutoBE, clone the repository and run the playground application locally. This allows you to chat with AutoBE's AI agents, manage multiple sessions, and use various LLM providers including local models like qwen3.5-397b-a17b.

After installation, the playground will be available at http://localhost:5713. You can interact with AutoBE through a chat interface - simply describe what you want to build, and AutoBE will generate the backend application for you.

Here's an example conversation script that guides AutoBE to create an "Economic/Political Discussion Board":

Requirements Analysis: "I want to create an economic/political discussion board. Since I'm not familiar with programming, please write a requirements analysis report as you see fit."
Database Design: "Design the database schema."
API Specification: "Create the API interface specification."
Testing: "Make the e2e test functions."
Implementation: "Implement API functions."

Compilation Success Dashboard

The playground includes a replay feature at http://localhost:5713/replay/index.html where you can view chat sessions from the AutoBE development team's testing and benchmarks.

Documentation Resources

Find comprehensive resources at our official website.

🏠 Home

🙋🏻‍♂️ Introduction
📦 Setup
🔍 Concepts

📖 Features

🔗 Appendix

How AutoBE Works

flowchart
subgraph "Backend Coding Agent"
  coder("Facade Controller")
end
subgraph "Functional Agents"
  coder --"Requirements Analysis"--> analyze("Analyze")
  coder --"ERD"--> database("Database")
  coder --"API Design"--> interface("Interface")
  coder --"Test Codes" --> test("Test")
  coder --"Main Program" --> realize("Realize")
end
subgraph "Compiler Feedback"
  database --"validates" --> prismaCompiler("<a href="https://github.com/wrtnlabs/autobe/blob/main/packages/interface/src/database/AutoBeDatabase.ts" target="_blank">Prisma Compiler</a>")
  interface --"generates" --> openapiCompiler("<a href="https://github.com/wrtnlabs/autobe/blob/main/packages/interface/src/openapi/AutoBeOpenApi.ts" target="_blank">OpenAPI Compiler</a>")
  test --"analyzes" --> testCompiler("<a href="https://github.com/wrtnlabs/autobe/blob/main/packages/interface/src/test/AutoBeTest.ts" target="_blank">Test Compiler</a>")
  realize --"compiles" --> realizeCompiler("TypeScript Compiler")
end

AutoBE follows a waterfall methodology to generate backend applications, with 40+ specialized agents handling each phase. The agents work in coordinated teams throughout the development process.

Each waterfall stage includes AI-friendly compilers that guarantee type safety of the generated code. Rather than generating code directly, AutoBE's agents first construct language-neutral Abstract Syntax Trees using predefined schemas. Each AST node undergoes validation against type rules before any code generation occurs, catching structural errors at the conceptual level rather than during compilation.

This approach is designed to ensure that the final generated TypeScript and Prisma code is 100% buildable. Based on our testing with multiple example projects including e-commerce platforms, discussion boards, and task management systems, AutoBE maintains this compilation guarantee across diverse application types.

To illustrate this process, here are the phase outputs from our "Economic/Political Discussion Board" example:

Requirements Analysis: Report
Database Design: Entity Relationship Diagram / Prisma Schema
API Specification: API Controllers / DTO Structures
E2E Test Functions: test/features/api
API Implementations: src/providers

Benchmark

AutoBE includes an automated evaluation pipeline that scores generated backends across 13+ LLM models and 4 project types. View the live results at autobe.dev/benchmark.

The benchmark evaluates compilation correctness, documentation quality, requirements coverage, test coverage, logic completeness, API completeness, and AI agent analysis (security, hallucination, code quality). Each model is scored 0-100 with grades A-F.

| Model | Todo | Reddit | Shopping | ERP | Avg | |-------|------|--------|----------|-----|-----| | glm-5 | 88 (B) | 87 (B) | 82 (B) | 87 (B) | 86 | | claude-sonnet-4.6 | 87 (B) | 85 (B) | 72 (C) | 85 (B) | 82 | | gpt-5.4-mini | 89 (B) | 87 (B) | 74 (C) | 78 (C) | 82 | | qwen3-coder-next | 86 (B) | 76 (C) | 75 (C) | 88 (B) | 81 | | qwen3.5-27b | 88 (B) | 81 (B) | 77 (C) | 78 (C) | 81 | | minimax-m2.7 | 90 (A) | 71 (C) | 77 (C) | 79 (C) | 79 | | gpt-5.4 | 79 (C) | 78 (C) | 79 (C) | 80 (B) | 79 |

Running the Benchmark

# All models (default)
corepack pnpm estimate

Autobe

Install / Use

README