mr-autopsy 0.1.1

Turn raw logs into incident postmortems — zero setup, fully local.

🔬 autopsy

Turn raw logs into incident postmortems — zero setup, fully local.

autopsy ingests logs from multiple services, correlates them by timestamp, detects anomalies, and outputs a structured incident postmortem — all from your terminal.

No agents. No cloud. No account. Just logs in, postmortem out.

---

🎬 Demo

$ autopsy --file api.log --file worker.log --file nginx.log

╭─────────────────────────────────────────╮
│ 🔬 autopsy  — incident postmortem generator │
╰─────────────────────────────────────────╯

📂 Reading: api.log       ✓ 16 events  (api)
📂 Reading: worker.log    ✓ 11 events  (worker)
📂 Reading: nginx.log     ✓ 11 events  (nginx)

⏱  Building timeline...
🔍 Running anomaly detection...
   ⚠  9 anomalies detected

╭─────────────── 📊 Summary ───────────────╮
│ Total events   38                        │
│ Errors         15  (39.5%)               │
│ Warnings        5                        │
│ Services       api, nginx, worker        │
│ Window         14:00:00 → 14:10:25       │
╰──────────────────────────────────────────╯

╭──────────────── 🚨 Anomalies ────────────────╮
│ 14:03:00  CRITICAL  error_spike              │
│           Error rate jumped to 93.3%         │
│                                              │
│ 14:03:01  ERROR     cascade                  │
│           worker erroring 14s after api      │
│                                              │
│ 14:03:10  WARNING   silence                  │
│           No logs from nginx for 430s        │
╰──────────────────────────────────────────────╯

✅ Done.

---

✨ Why autopsy?

Every observability tool assumes you have a budget, a team, and weeks to set up agents.

autopsy assumes you have a terminal and a log file.

	autopsy	Datadog	Incident.io
Setup time	30 seconds	Days	Weeks
Cost	Free	$$$	$$
Agents required	❌	✅	✅
Cloud account	❌	✅	✅
Logs leave machine	❌ Never	✅	✅
Works offline	✅	❌	❌
Works in air-gapped envs	✅	❌	❌

---

🚀 Install

pip install mr-autopsy==0.1.0

Requires Python 3.11+

---

⚡ Quick Start

# Analyze a single log file
autopsy --file app.log

# Multiple services at once
autopsy --file api.log --file worker.log --file nginx.log

# Filter to your incident window
autopsy --file api.log --from "2026-08-04 14:00" --to "2026-08-04 16:00"

# Read from a Docker container
autopsy --docker my-api --since 2h

# Pipe from kubectl
kubectl logs my-pod --since=2h | autopsy --stdin --stdin-service api

# Save a Markdown postmortem
autopsy --file api.log --output markdown --save ./postmortem-2026-08-04.md

---

📋 All Options

Options:
  -f, --file TEXT            Log file(s) to analyze
  -s, --service TEXT         Override service name (matches --file positionally)
  -d, --docker TEXT          Docker container name(s)
      --since TEXT           Docker --since value e.g. 2h, 30m
      --stdin                Read from stdin
      --stdin-service TEXT   Service name for stdin input  [default: stdin]
      --from TEXT            Start of time window (YYYY-MM-DD HH:MM:SS)
      --to TEXT              End of time window   (YYYY-MM-DD HH:MM:SS)
  -o, --output TEXT          Output format: terminal | markdown | json
      --save TEXT            Save report to file
      --bucket INTEGER       Bucket size in seconds for rate analysis [default: 60]
      --help                 Show this message and exit.

---

📁 Supported Log Formats

Format is auto-detected — you never need to specify it.

JSON (pino, winston, structlog, zerolog)

{"time":"2026-08-04T14:03:05Z","level":"error","msg":"DB timeout","service":"api"}

logfmt

ts=2026-08-04T14:03:05Z level=error msg="job failed" service=worker

Plaintext (nginx, syslog, any custom format)

2026-08-04 14:03:05 ERROR upstream returned 502 Bad Gateway

---

🚨 Anomaly Detection

autopsy runs three detectors on every analysis:

1. Error Spike

Detects when the error rate suddenly jumps above baseline.

14:00 ░░░░░░░░  2%   normal
14:01 ░░░░░░░░  3%   normal
14:02 ░░░░░░░░  5%   normal
14:03 ████████ 93%   ← CRITICAL: error spike detected

2. Cascading Failure

Detects when Service B starts failing shortly after Service A — suggesting an upstream dependency failure.

14:03:01  api     → first ERROR
14:03:05  nginx   → first ERROR  (4s later  → cascade detected)
14:03:15  worker  → first ERROR  (14s later → cascade detected)

3. Service Silence

Detects when a service stops emitting logs unexpectedly.

14:03:10  nginx  last log before silence
              ↕  430 seconds — no logs
14:10:20  nginx  logs resume

---

📄 Output Formats

Terminal (default)

Rich colored output with summary panel, anomaly table, and top errors.

Markdown

autopsy --file api.log --output markdown

Outputs a complete postmortem document:

# 🔬 Incident Postmortem
> Generated by autopsy v0.1.0

## 📊 Summary
| Window   | 2026-08-04 14:00:00 → 14:10:25 UTC |
| Duration | 10m 25s                             |
| Errors   | 15 (39.5%)                          |

## 🚨 Anomalies Detected
### 1. 🔴 Error spike detected
...

## 🕐 Event Timeline
...

## ✅ Action Items
- [ ] Identify root cause
- [ ] Add alerting for recurrence

JSON

autopsy --file api.log --output json

Machine-readable output for scripting or integrations.

---

🏗️ Architecture

Ingest → Parse → Correlate → Detect → Report

Layer	Job
Ingest	Read raw lines from files, Docker, or stdin
Parse	Convert each line into a structured LogEvent
Correlate	Merge all services into one sorted timeline
Detect	Find error spikes, silences, and cascades
Report	Render Markdown, JSON, or terminal output

Every layer has a single input and output type — swap any layer independently without touching the rest.

---

🗺️ Roadmap

• v0.1 — File ingestion, auto-detect format, timeline, anomaly detection, Markdown report
• v0.2 — Latency anomaly detection, HTML report output
• v0.3 — Custom log format config (.autopsy.yaml)
• v0.4 — Local LLM root cause summary via Ollama (offline AI, no API key)
• v1.0 — Full docs site, 80%+ test coverage, GitHub Actions CI/CD

---

🤝 Contributing

Contributions are welcome! Here's how to get started:

# Clone the repo
git clone https://github.com/yourusername/autopsy
cd autopsy

# Create virtual environment
uv venv
source venv/bin/activate

# Install with dev dependencies
uv pip install -e ".[dev]"

# Run tests
python -m pytest tests/ -v

Please open an issue before submitting a large PR — let's discuss the approach first.

See CONTRIBUTING.md for full guidelines.

---

💬 FAQ

Does autopsy send my logs anywhere? Never. Everything runs locally on your machine. No network calls are made.

What Python version do I need? Python 3.11 or higher.

Can I use this in CI/CD? Yes. Use --output json for machine-readable output and --save to persist the report as an artifact.

My log format isn't being detected correctly. What do I do? Open an issue with a sample (sanitized) log line and we'll add support. Format detection is a moving target and community samples help a lot.

Will you add real-time monitoring? No. autopsy is intentionally a post-incident tool. Use Prometheus, Grafana, or Datadog for real-time monitoring. Use autopsy after the incident to understand what happened.

---

📝 License

MIT © Tharun

---

Built for the developer who gets paged at 2am with nothing but a terminal.