Noise → Narrative

Gigabytes of logs → kilobytes of AI-ready insights.
Pattern and anomaly log extraction for AI and humans.

AI quick start

Connect Logoscope to Claude Desktop and analyze logs with natural language.

$ claude mcp add logoscope -- npx -y @probelabs/logoscope@latest mcp

Or manually add to ~/.claude/mcp.json (JSON config):

{
  "mcpServers": {
    "logoscope": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@probelabs/logoscope@latest", "mcp"]
    }
  }
}

Then ask Claude:

"Analyze these logs: /var/log/app.log"

Claude will automatically use fast triage analysis first, then deep analysis if needed.

CLI quick start

Two-step workflow: start with --triage for fast insights, then deep analysis.

$ # Step 1: Quick triage (fast anomaly detection)
$ npx -y @probelabs/logoscope@latest --triage logs/app-*.log

$ # Step 2: Full analysis (if anomalies found)
$ npx -y @probelabs/logoscope@latest logs/app-*.log --out insights.json

$ # Streaming logs
$ kubectl logs -n prod deploy/web --since=30m | npx -y @probelabs/logoscope@latest --triage
$ journalctl -u api -o json | npx -y @probelabs/logoscope@latest --out insights.json

Use --triage first for fast results, then run without it for comprehensive analysis. Accepts multiple files, globs, or standard input.

Anomalies

High‑signal items first: temporal spikes and new/rare patterns.

  "anomalies": {
    "temporal_anomalies": [
      {
        "type": "FrequencySpike",
        "template": "ERROR Database connection failed: <*>",
        "at": "2025-09-05T14:20:00Z",
        "value": 127,
        "baseline": 18
      },
    ],
    "pattern_anomalies": [
      {
        "type": "NewPattern",
        "template": "WARN retry=<*> exceeded for op=<*>",
        "first_seen": "2025-09-05T14:19:45Z"
      },
      {
        "type": "RarePattern",
        "template": "Payment declined: <*>",
        "frequency": 0.0031
      }
    ]
  },

Patterns

Top templates with severity, trend, and examples.

  "patterns": [
    {
      "template": "ERROR Database connection failed: <*>",
      "frequency": 0.0234,
      "total_count": 11700,
      "severity": "ERROR",
      "temporal": { "bursts": 2, "largest_burst": "2025-09-05T14:20:00Z", "trend": "increasing" },
      "examples": ["ERROR Database connection failed: timeout"]
    },
    {
      "template": "INFO user=<*> op=login ip=<IP>",
      "frequency": 0.1542,
      "total_count": 77100,
      "severity": "INFO",
      "temporal": { "bursts": 0, "largest_burst": null, "trend": "steady" },
      "examples": ["INFO user=9214 op=login ip=10.5.22.18"]
    },
    {
      "template": "level=info op=logout user_id=<*> status=<*>",
      "frequency": 0.0875,
      "total_count": 43750,
      "severity": "INFO",
      "temporal": { "bursts": 0, "largest_burst": null, "trend": "steady" },
      "examples": ["{\"level\":\"info\",\"user_id\":9876,\"op\":\"logout\",\"status\":\"ok\"}"]
    }
  ]

The Problem

Production is down. Your logs are gigabytes of noise: thousands of identical lines, buried anomalies, and patterns too complex for grep. Uploading to AI fails — context windows overflow, tokens waste on repetition, and temporal patterns vanish in copy-paste.

How It Works

Logoscope uses the Drain algorithm to extract log templates, replacing variables with wildcards. We identify anomalies (new patterns, frequency spikes, numeric outliers), track temporal dynamics (bursts, gaps, trends), and compress millions of lines into kilobytes of structured JSON.

What You Get

A complete analysis that fits in AI context: pattern statistics with severity and trends, temporal anomalies with exact timestamps, schema changes and field evolution, plus queryable access to raw logs. Use --triage for rapid anomaly detection, then deep-dive when needed.

Real Questions, Real Answers

"What spiked at 14:20?" • "Show new error patterns from today" • "Which fields changed before the outage?" • "Extract payment failures with context" — Logoscope makes logs answerable.

Noise → Narrative

Anomalies

Patterns

Part of the Probe Ecosystem