BAML

BAML 是一门专门写给 LLM 的小语言：把提示词和返回结构写在一起，编译成各语言可直接调用的类型化函数，模型返回不合 schema 时自动带着错误信息重试，不用手写一堆 JSON 解析和兜底。走的是代码生成的路子，Python、TypeScript 等项目里能拿到补全友好的客户端，适合要把模型输出当普通函数返回值用的后端。 🔗 文档: https://docs.boundaryml.com | GitHub Stars: 9.4k+

BAML is a small language written specifically for LLM calls: prompts and the expected return shape live in one place, compiled into typed functions callable from your language, with automatic retries that feed the schema error back to the model instead of hand-written JSON parsing and fallbacks. It works by code generation, so Python and TypeScript projects get completion-friendly clients, which suits backends that want model output to behave like an ordinary function return value. 🔗 Docs: https://docs.boundaryml.com | GitHub Stars: 9.4k+