Windows BSOD and Linux crash dump analysis, without the debugger console.
Upload a Windows or Linux crash dump, or a macOS crash report, and get a root-cause report back — faulting module, bugcheck code, stack, and a plain-language summary. We read the header, run the right command set, and have AI read the output. Cases the automated passes can't settle go to an engineer, and so far in the pilot that's been most of them; we'll publish the automated share once we've measured it.
How it works
Tier 1 — automated read
A fixed set of WinDbg/CDB/KD or gdb/crash commands runs against your dump (for a macOS report, a fixed set of reads of the report itself) and a model reads the output. Only a case the rules say is clear-cut is sent back without a human.
Tier 2 — deeper pass
Ambiguous cases get a second, longer pass: a wider command set (for macOS, more of the same report), checked for agreement with the first pass.
Tier 3 — engineer review
Cases neither pass can settle go to an engineer, and so does a crashing driver we can't identify. Heap corruption and memory-corruption exceptions do too, unless the deeper pass names the module that caused them (Windows) or a Linux application core's report names the tool that finds the earlier bad write. No false confidence. The Solo plan doesn't include engineer review: its reports always come from the automated passes, and a report they weren't confident in says so.
Built for MSPs, software vendors, IT admins and developers
- Windows and Linux. Windows user- and kernel-mode mini, kernel and full memory dumps, Linux process cores and kdump vmcores, all through the same pipeline. We detect which from the file header, so you don't have to know.
- Linux is in early access. A clear-cut Linux report comes straight back, like a Windows one, and anything the automated passes can't settle goes to an engineer (except on the Solo plan). To have an engineer review a dump before you get the report (not on the Solo plan), tick the box when you upload it, or send
human_review=truethrough the API. - macOS crash, panic and hang reports. App crash reports (.ips), kernel panics and spindump spin or hang reports. macOS writes them itself, already symbolicated, so there's no debugger and no worker: we read the report on our own servers. Our collection script lists the last week's reports and asks before sending any.
- Redaction before any AI sees it. Debugger output is scrubbed for secrets before it reaches the automated analysis. See what each step sends to which model and who processes what, and where.
- An audit trail, not a black box. Every command run against your dump is logged — never the output, only what was asked.
- Want to read a dump yourself first? Our crash dump analysis guides cover WinDbg, gdb and common stop codes such as 0x133.
- Browser or API. Upload in the web app, or send dumps from your own tooling through the API.
- Webhooks for your pipeline. Get notified the moment a report is ready instead of polling.
- Open-source capture scripts and agents. dumptriage-tools (MIT) turns on crash dump capture on Windows and Linux, and its agents send new dumps for you, by policy: always, ask first, or by rule. Download them.
Five design-partner places
Three months free, with human review included, in return for feedback. Or create an account for a free monthly allowance while we're in the pilot, or talk to us first if you need higher limits or an agreement in place.
See the design-partner offer