lapp

lapp

Find the problem in your logs, not the noise.

Drop in a log file and lapp shows you where to look first: errors that suddenly spiked and patterns that never appeared before. Click any of them to see the lines behind it, or hand it to Claude Code or Codex to investigate. Every claim in the answer links to the pattern and lines behind it.

To get there, lapp groups thousands of repetitive lines into a few named patterns and filters out the noise (health checks, heartbeats, routine requests). On a 2,000-line Apache sample log, that's 6 patterns, 70% of lines filtered, and one spiking error at the top of the list.

Your logs stay on your machine except in two cases you opt into. If you set an OpenRouter key, lapp sends a few sample lines from each pattern to an LLM to name the pattern; without a key, patterns are shown as their raw text. If you click Investigate, your Claude Code or Codex reads the filtered patterns, and like any session of theirs, what it reads goes to that agent's provider.

Quick start

lapp app.log                                    # a file
kubectl logs deploy/api | lapp --name api.log   # a pipe

lapp reads plain text, NDJSON, and Grafana Explore exports (Download as CSV or JSON). It finds the patterns and opens them in your browser. To pull logs from GCP Cloud Logging or VictoriaLogs, use Import logs in the page. A VictoriaLogs import splits logs by stream into one service each, named by the stream's service.name, service, app or job label, so each service keeps its own noise rules. VictoriaLogs credentials come from VICTORIALOGS_BEARER_TOKEN, or VICTORIALOGS_USERNAME and VICTORIALOGS_PASSWORD, in the environment or .env where lapp runs; the page never asks for them.

If VictoriaLogs is only reachable through Grafana, point --url at the Grafana data source proxy and use a Grafana service account token:

If a gateway such as Cloudflare Access sits in front, add its headers with VICTORIALOGS_HEADERS (Name: value, separated by ;), preferably in .env so the values stay out of shell history:

VICTORIALOGS_BEARER_TOKEN=<grafana-service-account-token> \
VICTORIALOGS_HEADERS='CF-Access-Client-Id: <id>; CF-Access-Client-Secret: <secret>' \
  lapp workspace import victorialogs --topic prod --since 30m \
  --url https://grafana.example.com/api/datasources/proxy/uid/<datasource-uid> \
  --query 'service.name:postgres_logs'
lapp workspace discover --topic prod-postgres-logs

1. Start here. A short list of the patterns most likely to be the problem: steady errors first, then spiking errors, other errors, warnings, other spikes, and patterns that are new since the last import of the same service.
2. Investigate. Click Investigate on any of them. Your local Claude Code or Codex reads the filtered patterns and explains the likely cause, citing the patterns and times it relied on.
3. Hide noise. Patterns that look like noise are tagged Likely noise, with the reason on hover. Click Hide or Not noise on the row, or hide any other pattern from its row. The share hidden and the lines left update as you go.
4. Keep the decisions. Every Hide, Not noise and Unhide is saved for the service right away, and lapp applies it the next time you open logs from the same service.
5. Take the result with you. Export an investigation summary (Markdown) for an issue, or export the filtered logs. From the command line:

lapp filter app.log > clean.log          # the log without lines hidden as noise
lapp summary app.log --json              # patterns, spikes and noise marks for scripts and coding agents

lapp never changes your original log files. Filtering only affects what you see and what you export.

The Logs volume chart shows lines over time when at least half of them have a time. Each pattern keeps its color in the chart and the list across imports of the same service. If a CSV file has no time lapp can read, choose its Time column in the chart card; lapp remembers it for the service. From then on the CSV header row is not counted as a log line, and exported lines (including lapp filter) leave it out.

Install

git clone https://github.com/STRRL/lapp && cd lapp
(cd frontend && pnpm install)
make build
export PATH="$PWD/output:$PATH"

Needs Go 1.26+, Node.js and pnpm. Optional: OPENROUTER_API_KEY for readable pattern names. Before anything is sent, lapp shows you exactly which lines will go to OpenRouter. Anything that looks like an email, IP address or token is replaced with <email>, <ip> or <token> by default.

Commands

The earlier lapp workspace … commands still work. The previous web UI is at /classic.

Contributing

make dev      # rebuild and start the web UI on 127.0.0.1:8080
make test     # unit + integration tests
make check    # format, lint, type-check, build, unit tests

Integration tests use the Loghub datasets: LOGHUB_PATH=/path/to/2k_dataset make integration-test. Design notes are in docs/ and CONTEXT.md.

License

MIT