Our Work
We're rekindling the dream of personal computing by making reliable software creation accessible to all.
Our first product, Sculptor, is a UI for running parallel coding agents safely in containers, instantly testing and merging changes, and getting suggestions and fixes for code as it's written.
Our research focuses on improving our ability to verify the correctness of LLM-generated code, so that anyone can trust the software they make is reliable and robust.
Product Preview
Sculptor: the missing UI for parallel coding agents
We rebuilt Sculptor from the ground up, thanks to your feedback. Now you can run multiple Claude Code agents in safe containers and instantly see their changes in action.
Read more
Our Functions
Product: Sculptor is an agent environment that helps engineers craft high-quality software by catching and fixing issues as they code. Over time, we hope to extend these capabilities to enable more and more people to build powerful, robust software that only teams of skilled engineers are able to build today.
Research: Our research focuses on building trust in LLM-generated code and agent outputs. There are two main thrusts towards that goal: (1) creating high quality datasets of complex and realistic coding problems, and (2) training verification models that validate code correctness.
Infrastructure: Our infrastructure group develops tools to streamline research and product development — from agent debugging interfaces to hyperparameter optimizers like CARBS.
Policy: Our policy work seeks to safeguard individual rights in an increasingly automated world, and ensure a fair and broad societal distribution of AI's benefits.
