Principles

Meet you where you are

• Ergonomics that feels like python but reads like a spreadsheet or pure formulas.
• Assumption tables respect data however it turns up. You dont have to have clean data, or do ETL from other system outputs
• Vector shimming. Performance is much better if you use vectors, but that happens by default without you worrying about it.
• Excel - We've tried to make it feel like you're writing a spreadsheet, but with the power of a programming language. There's also a lot of resources about how to convert models from excel, and familar function support.

Design for AI

• LLM native, great docs, great error messages built to be using (and tested with) agentic loops.
• MCP server, free agent for helping you build models, answer questions and help you debug.

Default fast, nudge to that.

• Works quickly on your local machine on CPUs, scales with 0 extra effort to GPUs.
• Key benchmarks are:
• How easy it is to make a change and see the result quickly. No warmup, no JIT, no compiling, no recursion, just works.
• You'll be in that loop for a while, change=>test=>refine so make that loop as tight as possible
• All of the python ecossytem /libraries available (easy to integrate and extend) but with clear guidelines on how to integrate to keep performance fast

Amazing docs

• Every method for every function is documented, and runs examples that have verfiied output.
• Documentation taylored to actuarial use cases.
• "Recipes" for common actuarial use cases.
• All documentation is built to be used by AI to help build models.
• Even when there's an error, the error message itself is crafted in a way to teach and help understand what to do next.