комп

logs in, signal outcut Claude’s log tokens by 99.5%

комп turns 2,001 log lines into about 344 tokens, on your machine.

what Claude received11 lines
[komp] 2001 lines → 9 entries · 2026-11-18 10:00:00–10:08:13 · service=orders
levels: FATAL 1 · ERROR 200 · WARN 200 · INFO 1200 · DEBUG 400
10:00:00–10:08:12 INFO  msg="GET /healthz" status=200 duration=1–4ms ×400
10:00:00–10:08:12 INFO  msg="GET /api/orders/{1011–9971, 32+ values}" status=200 duration=8–239ms request_id={32+ distinct, e.g. 7432d1026706d7e8} ×300
10:00:00–10:08:12 DEBUG msg="cache lookup" key=order:{1046–9938, 32+ values} hit=true ×218
10:00:01–10:08:13 INFO  msg="POST /api/orders" status=201 duration=20–90ms request_id={32+ distinct, e.g. a846a32c3bb81e3c} ×400
10:00:03–10:08:11 DEBUG msg="cache lookup" key=order:{1011–9973, 32+ values} hit=false ×182
10:04:06–10:06:50 ERROR msg="query failed" err="dial tcp 10.0.3.17:5432: connect: connection refused" query=GetOrder request_id={2720d45f49fdae09|17c52efacc51bd68|1f11a290b9c6acb4|5ae4a9bd1c7624f3|f8e43f1b9cd4ebab|ee3253cd6ca2aaf6|+26+ more} ×200
10:04:06–10:06:50 WARN  msg="retrying database connection" attempt={1|2|3|4|5}/5 backoff=2–32s ×200
10:04:06–10:06:49 INFO  msg="GET /api/orders/{1273–9955, 32+ values}" status=503 duration=14–237ms request_id={32+ distinct, e.g. 1550cde897d582c4} ×100
10:08:13 FATAL msg="giving up: database unreachable after 5 attempts"

❯ orders went down. check outage.log with komp and tell me why, in two short sentences

● komp - compress_file (MCP)

path: "outage.log"

⎿ [komp] 2001 → 11 lines, ~75958 → ~312 tokens (estimate)

[komp] privacy check: passed, nothing needed removing

● Orders dropped because the database at 10.0.3.17:5432 became unreachable (connection refused) from 10:04–10:06, causing `GetOrder` queries to fail and `/api/orders/*` requests to return 503s. The app retried 5 times but gave up at 10:08:13 with a fatal "database unreachable" error.

❯ ▍

Recorded on 18 November 2026 with комп v0.1.0 and a test log from the repository. комп ran with networking blocked by the macOS sandbox. The prompt, tool call, result and answer are unedited; scrolling only sets the pace.
fewer tokens in this run
99.6%
network calls
0
macOS, Linux, Windows
3 systems
free and open source
MIT

Every log line you paste uses Claude’s tokens and context, and most lines repeat. комп merges them first, so Claude gets the patterns, counts, time ranges and error values without reading the same line hundreds of times.

spend tokens on the fixnot on repeated logs

Measured on the four test logs in the repository. Tokens are estimated at about 4 characters each.

  • Go · database outage99.5% fewer

    ~75,958 → ~344 tokens

  • Spring Boot · NullPointerException99.5% fewer

    ~41,216 → ~220 tokens

  • Node.js · card declines99.5% fewer

    ~59,552 → ~295 tokens

  • kubelet · CrashLoopBackOff99.2% fewer

    ~39,180 → ~318 tokens

all four logs~215,906 → ~1,177 tokens

fast on your machinelight on memory

The real release binary, measured on an Apple M5 with 10 cores and 16 GB of memory. Even at 100,050 lines it finished in 632 ms and used 0.80% of system memory.

to compress 2,001 lines
27 ms
peak memory on that log
15.8 MB
of 16 GB system memory
0.10%
single binary
2.0 MB
Time as logs growmilliseconds
0200400600800025k50k75k100k632 ms

log lines

Peak memory as logs growMB
050100150025k50k75k100k131.5 MB

log lines

Each test log
loglinestimepeak memoryof system memory
Go2,00127 ms15.8 MB0.10%
Spring Boot1,35020 ms15.7 MB0.10%
Node.js1,20021 ms14.0 MB0.09%
kubelet70020 ms14.9 MB0.09%
Growth numbers as a table
linessizetimepeak memoryof system
2,0010.3 MB26 ms15.8 MB0.10%
4,0020.6 MB38 ms17.6 MB0.11%
10,0051.4 MB75 ms25.1 MB0.15%
20,0102.9 MB136 ms39.7 MB0.24%
50,0257.2 MB319 ms75.0 MB0.46%
100,05014.5 MB632 ms131.5 MB0.80%

Measured 18 November 2026 on macOS 27.2.1 (arm64) with other apps running. Each number is the median of 41 runs, or 15 for the growth charts, which repeat the Go test log. Memory is peak resident memory. Results will differ on other machines.

one commandthen just ask

Add комп to Claude Code once. Copy some logs and ask what went wrong. Claude reads the summary, so the raw logs never enter the chat.

claude mcp add --scope user komp -- komp serve
  • compress_clipboardwhatever you copied
  • compress_filea log file on your machine
  • compress_kubectl_logskubectl logs --timestamps with your current kube context

five stepsall on your machine

  1. 01

    Read every line

    Each log line is split into its time, level, message and stack trace.

    2026-11-18T10:00:01Z ERROR db timeout
    time 10:00:01 · level ERROR · db timeout

2,001 log lines in, 12 lines out

thousands of lines ina few lines out

An orders service loses its Postgres connection, retries, then gives up.

before2,001 lines · ~75,958 tokens
2026-11-18T10:00:00.313000000Z time="2026-11-18T10:00:00Z" level=info msg="GET /healthz" status=200 duration=1ms service=orders
2026-11-18T10:00:00.469000000Z time="2026-11-18T10:00:00Z" level=info msg="GET /api/orders/5506" status=200 duration=70ms request_id=7432d1026706d7e8 service=orders
2026-11-18T10:00:00.826000000Z time="2026-11-18T10:00:00Z" level=debug msg="cache lookup" key=order:1106 hit=true service=orders
2026-11-18T10:00:01.154000000Z time="2026-11-18T10:00:01Z" level=info msg="POST /api/orders" status=201 duration=74ms request_id=a846a32c3bb81e3c service=orders
2026-11-18T10:00:01.324000000Z time="2026-11-18T10:00:01Z" level=info msg="GET /healthz" status=200 duration=3ms service=orders
2026-11-18T10:00:01.686000000Z time="2026-11-18T10:00:01Z" level=info msg="GET /api/orders/6925" status=200 duration=155ms request_id=62179273c8eb5bb6 service=orders
2026-11-18T10:00:02.007000000Z time="2026-11-18T10:00:02Z" level=debug msg="cache lookup" key=order:5374 hit=true service=orders
2026-11-18T10:00:02.312000000Z time="2026-11-18T10:00:02Z" level=info msg="POST /api/orders" status=201 duration=41ms request_id=75ec87a171ac826a service=orders
2026-11-18T10:00:02.516000000Z time="2026-11-18T10:00:02Z" level=info msg="GET /healthz" status=200 duration=4ms service=orders
2026-11-18T10:00:02.767000000Z time="2026-11-18T10:00:02Z" level=info msg="GET /api/orders/8517" status=200 duration=44ms request_id=8478dcb74f21345d service=orders
2026-11-18T10:00:03.069000000Z time="2026-11-18T10:00:03Z" level=debug msg="cache lookup" key=order:2040 hit=false service=orders
2026-11-18T10:00:03.316000000Z time="2026-11-18T10:00:03Z" level=info msg="POST /api/orders" status=201 duration=79ms request_id=8038a39d5e085396 service=orders
2026-11-18T10:00:03.505000000Z time="2026-11-18T10:00:03Z" level=info msg="GET /healthz" status=200 duration=3ms service=orders
2026-11-18T10:00:03.850000000Z time="2026-11-18T10:00:03Z" level=info msg="GET /api/orders/3646" status=200 duration=146ms request_id=0af03b971722f244 service=orders
2026-11-18T10:00:04.168000000Z time="2026-11-18T10:00:04Z" level=debug msg="cache lookup" key=order:8787 hit=true service=orders
2026-11-18T10:00:04.385000000Z time="2026-11-18T10:00:04Z" level=info msg="POST /api/orders" status=201 duration=87ms request_id=d669cbee3772a077 service=orders
2026-11-18T10:00:04.536000000Z time="2026-11-18T10:00:04Z" level=info msg="GET /healthz" status=200 duration=1ms service=orders
2026-11-18T10:00:04.867000000Z time="2026-11-18T10:00:04Z" level=info msg="GET /api/orders/1964" status=200 duration=66ms request_id=21a278f64f7fd633 service=orders
2026-11-18T10:00:05.185000000Z time="2026-11-18T10:00:05Z" level=debug msg="cache lookup" key=order:8062 hit=false service=orders
2026-11-18T10:00:05.443000000Z time="2026-11-18T10:00:05Z" level=info msg="POST /api/orders" status=201 duration=72ms request_id=e131ca3766e4d58e service=orders
2026-11-18T10:00:05.656000000Z time="2026-11-18T10:00:05Z" level=info msg="GET /healthz" status=200 duration=1ms service=orders
2026-11-18T10:00:05.919000000Z time="2026-11-18T10:00:05Z" level=info msg="GET /api/orders/2604" status=200 duration=20ms request_id=0275dff6c15c0c8e service=orders
2026-11-18T10:00:06.142000000Z time="2026-11-18T10:00:06Z" level=debug msg="cache lookup" key=order:7930 hit=false service=orders
2026-11-18T10:00:06.331000000Z time="2026-11-18T10:00:06Z" level=info msg="POST /api/orders" status=201 duration=44ms request_id=9611a11f5125227c service=orders
2026-11-18T10:00:06.511000000Z time="2026-11-18T10:00:06Z" level=info msg="GET /healthz" status=200 duration=2ms service=orders
2026-11-18T10:00:06.809000000Z time="2026-11-18T10:00:06Z" level=info msg="GET /api/orders/1651" status=200 duration=166ms request_id=2da86a78c49ea20e service=orders
2026-11-18T10:00:07.118000000Z time="2026-11-18T10:00:07Z" level=debug msg="cache lookup" key=order:2638 hit=true service=orders
2026-11-18T10:00:07.405000000Z time="2026-11-18T10:00:07Z" level=info msg="POST /api/orders" status=201 duration=47ms request_id=84b27b95e9093483 service=orders
2026-11-18T10:00:07.782000000Z time="2026-11-18T10:00:07Z" level=info msg="GET /healthz" status=200 duration=1ms service=orders
2026-11-18T10:00:08.122000000Z time="2026-11-18T10:00:08Z" level=info msg="GET /api/orders/3546" status=200 duration=77ms request_id=96a68f812d810a48 service=orders
2026-11-18T10:00:08.313000000Z time="2026-11-18T10:00:08Z" level=debug msg="cache lookup" key=order:8239 hit=false service=orders
2026-11-18T10:00:08.606000000Z time="2026-11-18T10:00:08Z" level=info msg="POST /api/orders" status=201 duration=21ms request_id=3241b4d419b1b673 service=orders
2026-11-18T10:00:08.846000000Z time="2026-11-18T10:00:08Z" level=info msg="GET /healthz" status=200 duration=4ms service=orders
2026-11-18T10:00:09.245000000Z time="2026-11-18T10:00:09Z" level=info msg="GET /api/orders/3532" status=200 duration=68ms request_id=55d05ad7853c1f76 service=orders
2026-11-18T10:00:09.604000000Z time="2026-11-18T10:00:09Z" level=debug msg="cache lookup" key=order:8541 hit=false service=orders
2026-11-18T10:00:09.832000000Z time="2026-11-18T10:00:09Z" level=info msg="POST /api/orders" status=201 duration=49ms request_id=706ca828bca03858 service=orders
2026-11-18T10:00:09.991000000Z time="2026-11-18T10:00:09Z" level=info msg="GET /healthz" status=200 duration=1ms service=orders
2026-11-18T10:00:10.293000000Z time="2026-11-18T10:00:10Z" level=info msg="GET /api/orders/8119" status=200 duration=96ms request_id=ad3c681d06bd2aa3 service=orders
2026-11-18T10:00:10.627000000Z time="2026-11-18T10:00:10Z" level=debug msg="cache lookup" key=order:5920 hit=false service=orders
2026-11-18T10:00:10.947000000Z time="2026-11-18T10:00:10Z" level=info msg="POST /api/orders" status=201 duration=72ms request_id=ac946dc59c0996da service=orders
… 1,961 more lines
after12 lines · ~344 tokens
[komp] 2001 lines → 7 templates, 8 entries · 2026-11-18 10:00:00–10:08:13 · service=orders
levels: FATAL 1 · ERROR 200 · WARN 200 · INFO 1200 · DEBUG 400
10:00:00–10:08:12 INFO  msg="GET /healthz" status=200 duration=1–4ms ×400
10:00:00–10:08:12 INFO  msg="GET /api/orders/{1011–9971, 32+ values}" status=200 duration=8–239ms request_id={32+ distinct, e.g. 7432d1026706d7e8} ×300
10:00:00–10:08:12 DEBUG msg="cache lookup" key=order:{1011–9973, 32+ values} hit={true|false} ×400
10:00:01–10:08:13 INFO  msg="POST /api/orders" status=201 duration=20–90ms request_id={32+ distinct, e.g. a846a32c3bb81e3c} ×400
10:04:06–10:06:50 ERROR msg="query failed" err="dial tcp 10.0.3.17:5432: connect: connection refused" query=GetOrder request_id={2720d45f49fdae09|17c52efacc51bd68|1f11a290b9c6acb4|5ae4a9bd1c7624f3|f8e43f1b9cd4ebab|ee3253cd6ca2aaf6|+26+ more} ×200
    e.g. msg="query failed" err="dial tcp 10.0.3.17:5432: connect: connection refused" query=GetOrder request_id=2720d45f49fdae09
10:04:06–10:06:50 WARN  msg="retrying database connection" attempt={1|2|3|4|5}/5 backoff=2–32s ×200
    e.g. msg="retrying database connection" attempt=1/5 backoff=2s
10:04:06–10:06:49 INFO  msg="GET /api/orders/{1273–9955, 32+ values}" status=503 duration=14–237ms request_id={32+ distinct, e.g. 1550cde897d582c4} ×100
10:08:13 FATAL msg="giving up: database unreachable after 5 attempts"
99.5% fewer tokens

cheaper than tailwithout losing the facts

Shell shortcuts use fewer tokens than the whole log, but they drop facts or barely shrink it. In the Go outage log, the 503 responses are logged as INFO, so grep misses all 100 of them. tail keeps only the final fatal error from the outage. комп keeps both, with the time range.

Measured on the four test logs in the repository. The 14 facts come from комп's own tests and комп was built against these logs, so expect smaller gaps on other logs. Tokens are estimated at about 4 characters each.

  • the whole log14 of 14 facts

    ~215,906 tokens

  • sort | uniq -c14 of 14 facts

    ~212,899 tokens

  • grep errors and warnings9 of 14 facts

    ~35,083 tokens

  • tail -n 10010 of 14 facts

    ~17,259 tokens

  • комп14 of 14 facts

    ~1,177 tokens

The 14 facts, one by one
logfactsort | uniq -cgreptailкомп
Godatabase connection refusedkeptkeptlostkept
Gofinal fatal errorkeptkeptkeptkept
Go503 responses during the outagekeptlostlostkept
Goretry attemptskeptkeptlostkept
Gotime the outage begankeptkeptlostkept
Spring BootNullPointerExceptionkeptkeptkeptkept
Spring Bootline of code that failedkeptlostkeptkept
kubeletCrashLoopBackOffkeptkeptkeptkept
kubeletfailed health probekeptlostkeptkept
kubeletprobe connection refusedkeptlostkeptkept
Node.jsfailed chargekeptkeptkeptkept
Node.jscard declined messagekeptkeptkeptkept
Node.jsline of code that failedkeptkeptkeptkept
Node.jsfraud-check upstreamkeptlostkeptkept
Similar tools
toolmade fortakesneedssecretsstated reduction
компLogsClipboard, files, piped input, kubectlOne binaryRemoved on your machine99.5% on its test logs
RTKOutput of many shell commandsCommands the agent runsOne binaryListed for RTK Pro60–90%
Log ReducerLogsFiles, piped input, clipboard through /logdumpNode.jsNot stated, but shortens IDs and tokens70–90%
log-essenceLogsFiles, piped input, Docker, journaldPython 3.11+Redacted before analysis50%+

Other tools' details come from their public project pages, checked on 18 November 2026. They were not run here. Each figure comes from different logs, so the percentages cannot be compared directly.

less to readthe facts stay

01 / 06

Warnings and errors list their actual values and keep their stack traces.

ERROR query failed err="dial tcp 10.0.3.17:5432: connection refused" ×200

your logs stayon your machine

комп makes no network requests. It checks your logs locally and removes secrets and personal data before Claude gets the summary.

your machine

Your logsclipboard, file or kubectl
Privacy checksecrets and personal data removed
комп2,001 lines → 12
Claudereceives only the summary

stop pasting logsinto harness

curl -fsSL https://raw.githubusercontent.com/samchinmaya/komp/master/install.sh | sh
claude mcp add --scope user komp -- komp serve