Explain a log with PostHog AI

When a log line doesn't tell you enough on its own, open it and switch to the Explore with AI tab. PostHog AI reads that one record and writes up what it means, what probably caused it, and what to do about it – useful at 2am, or when the log came from a service you don't own.

This is a feature of the log detail view, not a chat you hold with an assistant. It explains the record you have open. To find the record in the first place, use the facet rail, filter bar, and free-text search in the logs viewer, or query logs from your editor over MCP.

What you get back

The explanation comes back as a structured writeup rather than a paragraph of prose:

• Headline and severity – a one-line summary and an assessment of whether this is likely fine, may need attention, is likely an issue, or is likely critical.
• Impact summary – a short statement of what or who is affected.
• Probable causes – up to three ranked hypotheses, each with a confidence level and the reasoning behind it.
• Immediate actions – up to five suggested next steps, each tagged do it now, address soon, or can wait.
• Technical explanation – the longer analysis, for when you want to understand the mechanism rather than just clear the alert.
• Key fields – the fields worth your attention, like a trace ID or an error code, and why each one matters.

What it reads

PostHog AI only sees the log record you opened: its message body, log and resource attributes, severity, service name, trace and span IDs, and event name. It doesn't search your other logs, and it can't act on them.

Because the input is a single record, the quality of the explanation follows the quality of your logging. Structured attributes and a meaningful service.name give it far more to work with than a bare string – see logging best practices.

Before you use it

Your organization needs AI data processing approved, since the log record is sent to a model to generate the explanation. If your logs may carry sensitive values, turn on PII scrubbing first – it redacts at ingestion, so scrubbed fields never reach storage or the explanation.

Explanations are cached for an hour per log record, so reopening the same line is instant and doesn't re-run the model.

Related

• Find the log you want to explain with search and filters.
• Spot the shapes worth explaining in log patterns.
• See everything the logs viewer does in Use Logs in the web app.
• Make your logs easier to explain with logging best practices.