wtf is h5i-db?

h5i-db is an analytical database built specifically for quantitative finance and time-series data. It is written in Rust and runs embedded in your project, meaning there is no separate server to set up or manage. You install it as a command-line tool or a Python library, and it stores data in a single directory on your machine. The database focuses on speed for financial workloads. The developers report that on a benchmark of 20 million rows computing common financial rollups, it runs over 4.5 times faster than well-known data tools like DuckDB and Polars. It achieves this by storing data presorted by time, which lets it skip sorting large datasets during queries. It also caches metadata aggressively because data segments are immutable once written. A core feature is versioning. Every write creates a new committed version of the data, and you can read any past version instantly. This means if you ingest bad data, you can restore to a previous state with a single command. The system also supports point-in-time reads, which let you pin a query to a specific timestamp so your results never include data that arrived later. This matters in finance research, where accidentally using future data can invalidate a trading strategy. The project is designed with AI agents in mind. A single command returns a full summary of the database: table schemas, sizes, time ranges, and any pending operations. Mutations like deletions can be previewed before they are applied, and a data policy can reject malformed rows automatically. Query results can be capped to avoid overwhelming an agent's context window, with overflow rows spilled to files instead. Error messages include suggested next actions and typo corrections. The developers are clear about limitations. h5i-db is single-node and embedded, so it is not suited for distributed multi-terabyte warehouses or high-concurrency transactional workloads. It also assumes your data has a time column, since the entire design depends on a time index for its performance and features. The project is open source under the Apache 2.0 license.