Autobe

AutoBE is an open source AI agent that generates complete TypeScript backend servers from a natural language conversation. It works through five phases in order: it extracts requirements, generates and validates a Prisma database schema, produces an OpenAPI specification, writes an end to end test suite, and finally implements the server in NestJS. Its central promise is that every line compiles, because each phase passes through a compiler before the next begins. The tool suits developers who want a working backend rather than a code sketch. Validation runs at three tiers, the Prisma compiler for the schema, an OpenAPI compiler for the API specification and the TypeScript compiler for the code, and the agent uses AI function calling to repair what fails. Every generated backend ships with a fully typed client SDK that serves both frontend integration and the generated tests. The documentation covers an agent library with configuration, event handling and prompt histories, plus a WebSocket protocol for running it from a client application or a Node.js or NestJS server. The site publishes worked examples such as todo, community forum and shopping backends, with token usage and success rates for each supported model, including models from Anthropic, OpenAI, DeepSeek, Qwen and Moonshot. AutoBE is free to use. The source is published on GitHub under the AGPL version 3 license, and there is no paid plan or pricing page; you supply your own model API access, whose costs depend on the provider. The project is developed by Wrtn Technologies, and its public roadmap marks the alpha through delta releases as complete, with the epsilon release in progress.