AI Alert Explanations

Available on: Starter, Pro, Enterprise

When a beet has an open alert, a “Explain with AI” button appears on the beet detail page. Clicking it streams a Claude-powered analysis of the incident directly in the alert banner.

What the AI receives

Heartbeet sends the following context to Claude with every explanation request:

Context	Source
Beet name and description	Your beet configuration
Alert severity	Current alert record
Time the alert fired	`triggered_at` timestamp
Expected vs actual event count	Current alert record
Hourly baseline for this exact time window	`beet_baselines_hourly`
Last 10 alerts for this beet	`anomaly_alerts` history
Other beets on the same project that fired within ±30 minutes	Correlated alerts query

The AI does not have access to your application code, logs, or any external systems.

What the AI returns

A streamed plain-English explanation (3–5 short paragraphs) covering:

What this means in business terms — translates raw numbers into impact
Most likely causes — reasoned from time of day, correlated alerts, and historical pattern
Recommended immediate actions — specific next steps, not generic advice
Recurring vs one-off assessment — based on alert history pattern

Example output

Your order.placed beet received zero events between 14:00 and 15:00, against a Wednesday afternoon baseline of 42–51 per hour. This is the first complete silence in 68 days of data.

Critically, three other beets on this project — payment.webhook, cart.abandoned, and checkout.initiated — also went silent within the same 4-minute window. This strongly points to a shared upstream failure rather than an issue isolated to your order flow.

Check your API gateway and payment processor status for the 13:58–14:03 window. A deployment rollout, rate limit breach, or DNS change is the most likely cause given the simultaneous drop across all four beets.

This does not match your Monday morning slowdown pattern. Treat as a one-off infrastructure incident until proven otherwise.

Regenerating

Once the explanation loads, a Regenerate button appears below it. This re-runs the analysis with fresh context — useful if the situation has evolved since the alert fired.

Model

Explanations use claude-sonnet-4-6. Token usage is tracked per organisation and visible on the admin dashboard.

Plan gating

Free plan users see a prompt to upgrade rather than the Explain button. The plan check is enforced server-side in the API route.

Usage tracking

Every explanation request logs input and output token counts to ai_usage_log. The admin dashboard shows:

AI calls in the last 30 days and all time
Estimated spend (at $3/MTok input, $15/MTok output for claude-sonnet-4-6)
Link to the Anthropic billing console for actual credit balance