Boundary AI

Boundary AI is a powerful toolkit designed to streamline the work of AI engineers, making it easier to harness the potential of large language models.

At its core, Boundary AI uses a unique configuration language called BAML, or Basically, A Made-up Language, which greatly enhances the functionality of LLMs. This innovative tool allows AI engineers to transform complicated prompt templates into easily executable functions. This simplification not only makes the execution more straightforward but also reduces the likelihood of encountering common errors related to parsing and typing.

Working with an LLM through BAML is comparable to calling a regular function, making it instinctive and user-friendly. Plus, the toolkit allows for quick testing of new prompts across various integrated development environments (IDEs), notably with the help of BAML's VSCode Playground UI. This instant feedback feature is incredibly useful for engineers looking to refine their prompts efficiently.

Another valuable component of Boundary AI is Boundary Studio, which provides tools for monitoring and tracking each LLM function's performance over time. This feature helps engineers assess how well their implementations are working and identify opportunities for improvement.

Built primarily in Rust, BAML supports a variety of models, including those from OpenAI, Anthropic, Gemini, Mistral, and even user-uploaded models, with future plans to extend capabilities to non-generative models. When deploying with BAML, the toolkit generates code in either Python or TypeScript, ensuring versatility in development.

What sets BAML apart from other data modeling libraries is its strong focus on type safety and transparency, meaning prompts are never hidden or obscured. Users benefit from an integrated playground that accommodates any model and encourages experimentation. The BAML compiler and the VSCode extension are both freely available as open source, while premium features for monitoring and enhancing functions in Boundary Studio come at a cost.