Tafuta
Open a 2 GB, 27-million-row CSV in 2 ms. Search it in under a second.
- A concept, on invented data
- Desktop (Tauri / Windows, macOS, Linux)
The problem
Standard tools — spreadsheet apps, text editors, basic CSV viewers — run out of memory or become unusable with files above a few hundred megabytes. Gigabyte-scale CSVs and logs need something that never loads the whole file.
The idea
Tafuta (Swahili for search) is a Tauri v2 desktop app with a Rust engine that memory-maps the file, builds a parallel row index, and serves a small window of rows to a virtualised React table on demand. The file is never fully in memory.
What it does.
- Opens files of any size instantly, showing the first rows in 2–3 ms while the full index builds in the background
- Searches 2 GB of CSV in under 700 ms with plain text or regex, streaming matches as they arrive
- Handles correct CSV: quoted commas, escaped quotes, embedded newlines, CRLF, BOM, invalid UTF-8
- Follows growing log files live, with ERROR/FATAL in red and WARN in amber
What’s underneath.
The decisions that make it work, most of which no one sees.
The indexer and the parser agree by construction
A quote only opens a quoted field at the start of a field, matching csv-core's rule exactly. The byte scanner and the field splitter can never disagree on where a row ends, so there is no second pass to reconcile them.
Rust computes highlight ranges as UTF-16 offsets
get_rows returns each match's start and end as UTF-16 offsets — what JavaScript strings use — so the table never re-implements regex semantics in the frontend. Embedded newlines are shown as ↵ with length-preserving offsets.
Each engine owns its own thread pool
Sharing rayon's global pool let a cancelled search's chunks queue behind another engine's live work. Moving each engine to its own pool, and checking the cancel flag every 64 rows, brought cancel latency from 1.49 s to under 200 ms.
What it shows
This concept is built from this service. Each is something we can do for you.
Tell us what you’re building.
A 30-minute call is the quickest way to see if we’re the right fit. Or send us a message, whichever suits you.