Boundary is the company behind BAML, an open-source domain-specific language for building reliable LLM applications. With BAML, developers describe an LLM function declaratively, its name, input types, output schema, prompt template, and provider, in BAML source files. The BAML toolchain then generates type-safe clients for TypeScript, Python, Ruby, and other supported languages, with structured-output parsing, retries, streaming, and tracing built in.
The company was co-founded in 2023 by Vaibhav Gupta (CEO) and Aaron Villalpando. Gupta brings nearly a decade of engineering experience from D.E. Shaw, Google's augmented reality group, and Microsoft HoloLens, where he built predictive pipelines and real-time computer vision systems. Boundary participated in Y Combinator's Winter 2023 batch and has since raised seed financing from YC and a syndicate of developer-tools angels; the company has not disclosed all round details publicly.
BAML is open source and the company monetizes via a paid cloud and enterprise offering that adds hosted observability, evaluations, and team collaboration on top of the open language. The core CLI, VS Code extension, and language server are free to use, which fuels organic adoption in the developer community.
Developers reach for BAML when they want LLM calls to behave like real functions, typed inputs, typed outputs, deterministic parsing, and clean failure modes, rather than free-form Python strings. The language ships with a playground where developers can iterate on prompts and immediately see structured outputs validated against their declared schemas. BAML also supports parallel calls, retries, and built-in tool/function calling, which lets teams compose multi-step agentic pipelines without boilerplate.
The competitive landscape is split between low-level libraries (Instructor, Outlines, Marvin) that bolt structured output onto Python and higher-level platforms (LangChain, LlamaIndex, Vellum) that try to be full agent frameworks. BAML occupies the middle ground: a real DSL with a compiler, but lightweight enough to slot into any application without committing to a heavy framework. That positioning has produced strong organic search traffic for queries like 'BAML LLM', 'structured output LLM', and 'BAML vs Instructor'.
The key differentiator is that BAML is a language, not a library. Schema mismatches, retry policies, and provider switching are caught at compile time, which makes BAML especially attractive to teams shipping production AI features that have to keep working as models, prompts, and providers change.