BAML

Overview

BAML is a domain-specific language for LLM functions. Define your input and output types, write the prompt, and BAML generates typed client code in the language of your choice. The result is LLM calls that act like regular functions, with first-class testing, observability, and provider-agnostic deploy.

Best for

Best for
Teams who want LLM calls to feel like real typed functions

Use cases

• Define LLM functions once, use them from any language
• Get typed structured output with strong validation
• Run evals and tests on prompts the way you do on code
• Swap providers without rewriting the calling code

Notes

Why it matters

The promise is appealing: LLM calls as typed, testable, language-agnostic functions. BAML is the clearest articulation of that vision shipping today.

How teams use it in production

Define LLM functions in BAML, generate clients, call them like regular typed functions from your app. Run evals against test cases in CI like any other code change.

What to watch

Whether a DSL is the right abstraction, or whether annotated typed functions in the host language (Instructor, structured-output SDKs) win. The argument is live.