Hyperparam has developed Squirreling, a compact SQL engine tailored for the web, addressing the challenges of querying large LLM datasets directly in the browser. Unlike traditional SQL engines that rely on WebAssembly, which often results in slow performance and high memory usage, Squirreling is designed for optimal interactivity. At around 9KB, it has a minimal footprint, allowing for immediate startup and efficient memory management. The engine employs an async-native execution model, enabling real-time query results and a responsive user interface. Squirreling’s architecture incorporates late materialization and an AST-driven execution model. By delaying the computation of column values until they are needed, Squirreling conserves resources and accelerates performance. It distinguishes between streaming and buffered processing paths depending on the complexity of the query, ensuring that only essential data is handled at any stage. The engine also avoids the overhead of a separate planning phase, allowing for more efficient, incremental execution of SQL queries directly within the browser. The small size and async capabilities enable Squirreling to operate entirely client-side without dependencies on a backend service. Users can explore data interactively and leverage asynchronous data sources, including cloud-native formats like Parquet, without the need for complex setups. This approach fundamentally changes how users can engage with large datasets, offering a lightweight solution for real-time data analysis. The project is open-source and available on GitHub for further exploration and use.