logdelta

diff logs by meaning, not by bytes

What happened in the failing run that didn't happen in the good one?

Paste a known-good run and the run you're investigating. logdelta masks what always differs (timestamps, IDs, IPs, durations), groups lines into templates, and reports the templates that are new, gone or changed in frequency, and the values that flipped. This is the Rust library itself, compiled to WebAssembly: your logs stay in this tab.

Examples

Good runs baselines

Run to investigate target

Numbers are normally masked. Watch status or exit codes to catch value changes even when the log's shape stays the same. One JSON Pointer per line, such as /http/status.

Keep routes or services separate for every watched field. One JSON Pointer per line, up to four. A missing group key makes the watch incomplete; leave blank to pool records.

Catch an error that was already present but became common. Compare each value's share of logged observations within its group. Unobserved groups have unknown rates.

Options

Up to 8 baselines, 25 MiB per log after decompression and 50 MiB total. Use the CLI for larger logs.

Comparison