Benchmarks

How Kelora compares with specialized tools on common log tasks. Single-purpose tools are usually faster at their one job; Kelora's numbers vary with the task — built-in options are quick, scripts that run on every event are slower. These are single runs on one machine, so read them as rough orders of magnitude, and combine tools: rg in front of Kelora, jq or qsv behind it.

Your main job	Fastest tool	Kelora is worth it when
finding text	rg, grep	you need fields, scripts, or counts in the same run
filtering JSON	jq, Kelora	you also need counts, windows, or non-JSON input
reshaping JSON	jq	the logic is multi-step, or you need counts, windows, or mixed formats
CSV analytics	qsv, mlr	the input isn't CSV, or you need Rhai logic

Results

Single runs, wall-clock time, 100 000 lines unless noted. Lower is better.

Machine: a 2018 Mac mini (Intel Core i5-8500B, 6 cores, 16 GB RAM), macOS 15.7, Kelora 2.1.1 (c5d80e2b5), October 2026. Current hardware is considerably faster, so compare rows with each other rather than reading the absolute times. Tools: ripgrep 15.1.0, ugrep 7.8.4 (as grep), jq 1.6, mlr 6.15.0, qsv 8.1.1, angle-grinder 0.19.5, klp 0.77.0.

Task	Kelora	Others
find ERROR lines in plain text	0.18 s	grep 0.05 s, rg 0.05 s, angle-grinder 0.34 s, klp 1.91 s
extract three columns from text	0.63 s	awk 0.37 s, angle-grinder 0.34 s, klp 8.18 s
filter JSON by level	0.14 s	jq 0.95 s, angle-grinder 0.34 s, klp 4.07 s
filter JSON, add a computed field	0.71 s	jq 0.65 s, angle-grinder 0.22 s, klp 12.6 s
count errors by component	0.78 s	jq + sort + uniq 0.59 s, angle-grinder 0.21 s, klp + sort 3.32 s
500 000 JSON lines, filter and count	2.60 s; --parallel 0.81 s	jq 3.95 s, angle-grinder 1.01 s, klp 29.3 s
filter CSV, select columns	2.22 s	qsv 0.19 s, mlr 0.34 s

As a rule of thumb: built-in options like -l, --keep-lines, and --freq are several times faster than Rhai scripts that run on every event, and --parallel gave about 3× on six cores. Big Files, Many Files lists what makes a run faster.

Run them yourself

cargo build --release
./benchmarks/generate_comparison_data.sh
just bench-compare

The script writes one table per task to benchmarks/comparison_results/. Numbers vary with hardware and tool versions; compare tools on the same machine rather than across machines.