Search many log files as one timeline. Any format, real filters, and one flag to follow a single request across all your services.
npx @ragib.mrb/logsift logs/ --request req-211-05 13:23:05.000 INFO api.log request start reqId=req-2 method=POST url=/orders
11-05 13:23:05.100 INFO gateway.log proxy request_id=req-2 path=/orders status=502
11-05 13:23:05.500 DEBUG worker.log picked job for request req-2
11-05 13:23:06.000 ERROR worker.log charge failed for req-2
Error: card declined
at charge (/app/billing.js:42:11)
11-05 13:23:06.500 ERROR api.log payment failed reqId=req-2 status=502
Three services, three log formats, one story.
grepsees text, not time: lines from different files come out file by file, not in the order things happened.- Pretty-printers like pino-pretty format one stream; they don't merge files or follow a request.
- The great log tools (lnav, hl) aren't one
npxaway.
logsift reads every file you give it, works out each line's format, and merges everything by time as it streams. A million-line, 115 MB file is searched in about 3 seconds.
npm install -g @ragib.mrb/logsift # gives you the `logsift` command; or just: npx @ragib.mrb/logsift …Needs Node.js 22 or newer. No dependencies.
logsift api.log worker.log # several files, merged by time
logsift logs/ # every log file in a folder (and .gz files)
logsift "logs/**/*.log" # globs work on Windows too
kubectl logs my-pod | logsift -l warn # or pipe anything inAll filters must match.
| Option | What it keeps | Example |
|---|---|---|
-l, --level |
this level and above | -l warn |
-s, --since / -u, --until |
a time range | -s 15m, -s 14:03, -u "2026-11-05 14:30" |
-g, --grep |
text (any case), /regex/flags or re:regex; repeat to require several |
-g timeout, -g "re:fail|denied" |
-w, --where |
field filters: =, !=, >, >=, <, <=, ~ (contains) |
-w "status>=500" -w user.id=42 |
-r, --request |
every line that mentions one request, trace or correlation ID | -r req-2 |
-a, --around + --window |
what happened just before and after a moment | -a 14:03 --window 30s |
--request matches the whole ID only (so req-42 doesn't match req-420). It checks the text of each line and common ID fields such as reqId, requestId, traceId, correlationId and x-request-id.
Git Bash on Windows rewrites arguments that start with
/into file paths, which breaks/regex/. Usere:regexinstead.
| Option | What you get |
|---|---|
| (default) | readable, coloured lines with the time, level, file and fields |
--json |
one JSON object per line, for jq or another tool |
--stats |
counts by level and file, and the most common messages (numbers and IDs grouped) |
-n, --limit |
stop after n matching lines |
-f, --follow |
keep watching the files and print new matching lines |
--no-color |
plain text (also when NO_COLOR is set) |
Exit code: 0 when something matched, 1 when nothing did, 2 on errors, like grep.
| Format | Examples |
|---|---|
| JSON lines | pino, bunyan, winston, Elastic (ECS), most JSON loggers |
| logfmt | time=… level=info msg="gateway up" port=8080 |
| Plain text | 2026-11-05 14:03:01,123 ERROR …, [2026-11-05T14:03:01Z] [warn] …, nginx and syslog times |
- Times can be ISO strings, seconds or milliseconds; levels can be words or pino/bunyan numbers.
- Stack traces and other lines that continue an entry stay attached to it.
- Lines with no time keep their place right after the line before them.
- Plain-text times without a time zone are read as your local time.
import {
readEntries,
mergeByTime,
buildFilter,
expandInputs,
labelSources,
openStream,
} from '@ragib.mrb/logsift';npm install
npm test
npm run build && node dist/cli.js test/fixtures --request req-2Want to help? See CONTRIBUTING.md. Changes are listed in CHANGELOG.md.