AI error analysis

AI analysis classifies each new incident by severity and affected component, with a one-line summary — so you can triage at a glance. Bring your own model, or rely on the built-in heuristic.

AI analysis is available on the Pro plan and above.

Two modes

Built-in heuristic (always on for entitled plans). Without any configuration, error.page classifies incidents using a deterministic heuristic: severity is inferred from the error type, HTTP status, and blast radius; the component is inferred from the failing endpoint. It's fast, free of external calls, and always produces a result.

Bring your own model. Connect your own OpenAI, Anthropic, Google Gemini, or self-hosted model for richer classification. Your key is encrypted at rest (AES) and used only to categorise your project's errors — we never proxy it anywhere else.

Connecting a model

From AI Analysis in the dashboard:

1. Choose a provider (OpenAI, Anthropic, Gemini, or Custom).
2. Optionally name a model (e.g. gpt-4o-mini, claude-haiku-4-5, gemini-1.5-flash). Sensible defaults are used if you leave it blank.
3. Paste your API key. Only the last four characters are ever shown again.
4. For a custom self-hosted model, provide an HTTPS endpoint. It's validated against our SSRF protections (no private or reserved addresses).

When categorisation runs

Classification runs automatically on the first occurrence of each new incident. The result appears on the incident in the feed and on its detail page as a severity badge, a component label, a short summary and tags.

Reliability

If your model call fails, times out, or returns something unparseable, error.page silently falls back to the heuristic — an incident is never left uncategorised, and a provider outage never blocks ingestion.

Cost

Because you use your own key, model usage is billed by your provider, on your account. error.page adds no markup — you only pay your provider for the tokens spent.