Sparrow — Iceberg stops at the files. Sparrow starts at the serving index. I wrote the clients.
…and I wrote the clients.
The missing client toolkit for Apache Arrow Flight.
Flight moves columnar data at wire speed — but the client side of the ecosystem is thin.
Sparrow fills it: five clients, one protocol — the terminal,
the browser, the spreadsheet, your agents, and a diagnostic.
Terminal
sparrowCLI — Go, for humans and agents: catalog, SQL, Parquet/CSV export, a connection doctor, verifiable receipts
Browser
sparrowJS — TypeScript, on npm: Flight SQL over gRPC-web, streaming record batches into the page
Excel
sparrowXL — a native add-in streaming Arrow straight into the grid
Agents
sparrowMCP — any Flight endpoint speaks MCP (Model Context Protocol)
X-ray
Flight Inspector — point it at any Flight SQL server: conformance, catalog, wire anatomy
They work with other people's servers too — Dremio,
Spice.ai, InfluxDB Cloud — not just
ours. The CLI is validated against five vendors; the Inspector reaches them live.
Underneath, for the live demos: the
Sparrow serving engine — a file that feels like
a database. The serving index ships with the data; a server memory-maps the snapshot and
starts answering.
8 ms from a local file,
27 ms across the LAN.
Every demo on this site runs against its
public
endpoint — the energy data there is just the demo. The tools are the product.