Shorten the first incident pass
Turn one failing provider event into a ranked cause list before opening five docs tabs.
Free log auditor
Audit one failing AI API log.
Paste an OpenAI or Anthropic failure event. causely ranks likely causes, cites the vendor docs, and returns a safe patch to try.
Why this tool
Debug one provider failure without turning an incident into a research project.
Stay inside the supported scope
OpenAI and Anthropic 429, quota, model, payload, timeout, 500, and overload signals only.
Keep the fix reviewable
Every result includes evidence, vendor docs, and a small patch you can copy or download.
How it works
The route is live: input changes the answer, and bad input stays empty.
- 01
Paste a real OpenAI or Anthropic error event with status, endpoint, and request boundary.
- 02
Watch the readiness preview update as you type so you know what signal the tool can inspect.
- 03
Run the API-backed audit. Unsupported or empty input returns a plain error, not a populated result.
- 04
Review the ranked cause, copy the patch, download the report, or reset for the next log.
Live preview
See what the auditor can read.
The preview updates as you type and only reports detected fields. The ranked cause still comes from the audit API below.
updates live
This preview only checks visible fields. It does not produce a diagnosis or patch.
Readiness output
needs signalProviderNot detected
StatusMissing status code
SignalNo supported failure signal
EndpointOptional but useful
No populated result yet
Start with a real OpenAI or Anthropic error event. Empty input stays empty.
Benefits
A focused diagnostic loop for teams shipping customer-facing AI features.
Incident-first
Built for a failing trace in front of you, not broad observability setup.
Docs-linked
Suggestions point back to vendor guidance so engineers can verify the next move.
Low ceremony
No account or integration required for the free audit route.
Interactive audit workspace
Paste a log, run the audit, keep the result.
This is the real tool route. Empty, offline, or unsupported input returns an error instead of a filled result.
causely.megaloop.app/tools/log-audit
paste provider log, rank likely cause
local corpus validator
Ranked causes
docs + patchWaiting for a provider error log
Empty or unrelated text returns no populated audit. The prototype only ranks supported OpenAI and Anthropic failure signals.
Founding pilot
Need causely in the incident loop?
The paid pilot adds weekly incident review and usefulness labeling around real OpenAI and Anthropic traces.
Bring one recent failure pattern. We will follow up only if causely can stay inside the supported scope.
FAQ
Narrow scope keeps the free tool honest.
Does the free auditor store logs?
No secrets are needed. Paste only the provider error boundary you are comfortable sending to the audit endpoint.
What happens with unrelated text?
The tool refuses empty or unsupported input instead of inventing a realistic-looking diagnosis.
Can I use Python instead of TypeScript?
Yes. Pick the language toggle before running the audit and the patch follows that choice.
Is this a root-cause promise?
No. causely ranks likely causes and gives the fastest safe next step for an engineer to verify.