Researchers introduce Hyperparam, a suite of three lightweight JavaScript libraries totaling under 70 KB that enable AI agents and client-side applications to query structured data directly from cloud storage without traditional data warehouse infrastructure. The system achieves 300x performance improvements over DuckDB-WASM on certain queries by integrating language model-based text interpretation directly into SQL execution.
The emergence of client-side AI agents running in browsers and lightweight environments creates a fundamental mismatch with existing data infrastructure. Traditional data warehouses like Spark, Trino, and managed solutions require significant computational resources and network overhead—impractical for applications that must operate within constrained runtimes like browser tabs or per-turn agent sandboxes. Hyperparam addresses this architectural gap by delivering a JS-native query engine specifically designed for this constraint.
The research reflects a critical shift in how AI applications interact with data. As agent-based systems proliferate in production environments like Claude Code and Cursor, these tools increasingly need to analyze unstructured data—agent traces, reasoning chains, conversation logs—that standard SQL cannot effectively query without model inference in the execution path. The three-library approach (Hyparquet for Parquet file reading, Squirreling for LLM-native query execution, and Icebird for Apache Iceberg support) demonstrates thoughtful architecture, with async-native execution preventing expensive operations from firing unless downstream operators require results.
Performance metrics reveal tangible efficiency gains: 300x faster filtering and 192x faster sorting compared to browser-based alternatives, while reducing agent analyst task costs by two-thirds. These improvements matter because they directly affect operational expenses for AI service providers and enable smoother user experiences in real-time agent interactions.
The implications extend beyond engineering optimization. This work signals that data engineering practices must evolve to accommodate AI-native application design. As agents become integral to production systems and users expect seamless data analysis within chat interfaces, the traditional separation between application runtime and data processing infrastructure becomes untenable. Organizations building next-generation AI applications should recognize this infrastructure category as essential rather than optional.
- →Hyperparam provides three open-source JavaScript libraries under 70 KB enabling direct Parquet and Iceberg querying from object storage in client-side AI applications.
- →The system achieves 300x performance improvement over DuckDB-WASM on filter-bounded queries through async-native SQL execution and LLM-shaped user-defined functions.
- →Client-side AI agents require fundamentally different data infrastructure than traditional warehouses, prioritizing bundle size, JS-native distribution, and integrated model inference.
- →Cost reductions of two-thirds for agent analyst tasks demonstrate economic viability of the approach compared to conventional query engines.
- →Data engineering practices must evolve to support AI-native client applications where agents and humans operate in the same process with shared data access.