Debug with AI without pasting production data

Logs leak more than logs

A production log line rarely contains only the error. It contains the request that caused it: the user's email, their account id, the address they typed, sometimes the full payload. A stack trace can carry variable contents. A config file carries hostnames and connection strings.

Pasting that into a cloud AI tool is a disclosure of production data, even though it feels like a technical question rather than a data-sharing decision.

Redact the data, keep the shape

What makes a log useful to an AI is its structure: the sequence of events, the error types, the timing, the repetition. None of that requires the real values. Consistent labelled placeholders keep the correspondence — the same user id remains the same placeholder across every line — so a pattern that spans a hundred lines is still visible.

The app reads source code and developer config files directly, alongside plain text, so you can work with the file rather than copying fragments out of it.

Add your own internal terms

Internal hostnames, service codenames, unreleased feature flags and project names are sensitive in your organisation and invisible to any general PII detector. Add them as custom confidential terms and they get their own placeholders too.