ax

Why we included this project

Ax brings the DSPy approach to TypeScript, letting you define each LLM call as a typed signature backed by a validator like Zod, Valibot, or ArkType, then leaving prompt construction, parsing, validation, streaming, and retries to the library. That is a real relief when your team has been juggling raw prompt strings and brittle JSON parsing. It also includes agents with memory and delegation, typed flow graphs that support branching and caching, and optimizers such as few-shot bootstrapping and GEPA for improving output quality. The same core program model compiles to Python, Java, C++, Go, and Rust packages, so if your stack spans runtimes you can keep one design and a provider abstraction that covers OpenAI, Anthropic, Gemini, Mistral, and local endpoints without rewriting prompts.