Systopsy · Coming next

An autopsy for crashed servers.

Point Systopsy at a host that went down. It collects the evidence, rebuilds the timeline and tells you what killed it, in plain language with the log lines to prove it.

Request early access Source goes public at beta

The problem

When a production server crashes, the cause is almost always written down somewhere: the kernel ring buffer, journald, a service's error log, the OOM killer report. But it's spread across many files and timestamps, and the engineer reading it is under pressure. The first part of every incident is the same tedious grep-and-correlate work.

The approach

Systopsy automates that first part. A local collector gathers evidence and builds one ordered timeline. A rule engine flags well-known failure patterns. Then, if you choose, an AI model reasons over the reduced, redacted evidence and writes a ranked root-cause report with cited lines and concrete fixes.

systopsy

$ systopsy collect --since "2h ago"
✓ kern.log  ✓ journal  ✓ dmesg
✓ app logs  ✓ metrics  (3 secrets redacted)

$ systopsy analyze --ai none
rule OOM_KILL       mysqld  HIT
rule DISK_FULL      /var    ok
rule KERNEL_PANIC   -       ok
› exported evidence.md + prompt.txt

$ systopsy analyze --ai local
› model running on this machine
› nothing sent over the network
✓ report written to report.md

Choose where the analysis runs

Claude is the default and our best-supported option. You can also use another AI provider, run a model locally, or skip AI completely.

	Claude	Any AI API	Local LLM	No AI
Good for	Most users. Our primary, best-supported provider	Teams standardised on another provider	Air-gapped hosts and strict data policies	Quick triage, no AI subscription
Analysis	The model correlates the timeline, ranks likely causes and cites the exact log lines			Built-in rules for common failures plus a readable evidence report
Output	Root cause, confidence level, fix commands			Rule hits, timeline and a ready-made prompt to paste into any assistant
Data leaves your network?	Only the reduced, redacted evidence, and only after you confirm		Never
You need	A Claude API key (hosted tier planned)	An API key for your provider	A local model runtime, for example Ollama or LM Studio	Nothing

What it understands

Linux & kernel

OOM kills, kernel panics, hung tasks, segfaults and filesystem errors.

Services & databases

Failed systemd units, crash loops, database crashes and recovery, memory misconfiguration.

Resources

Memory, disk and I/O exhaustion, runaway processes and full filesystems.