aiinsights

AI Insights

Get plain-English summaries of recent log activity and track persistent patterns — generated by Workers AI running on Cloudflare's edge.

AI Insights

AI Insights analyzes your recent log activity and writes a plain-English summary of what happened — error patterns, active services, and anything worth investigating.

What you’ll need

Step 1: Open AI Insights

Click Insights in the left sidebar.

The page shows:

Step 2: Generate a summary

Click Generate Summary and choose the time window:

ScryWatch reads your event counts, error counts, and pattern frequencies for the selected window, then calls Workers AI (Llama 3.1 8B) to write a summary.

The summary typically looks like:

“In the last 15 minutes, 2,341 events were ingested across 3 services. The api-gateway emitted the most events (1,204). There were 18 errors, primarily from the payment-service pattern stripe charge failed: {code}. This is a new pattern that first appeared 4 minutes ago and may be worth investigating.”

Note: Summary generation takes 3–8 seconds depending on load. The result is stored and can be retrieved without regenerating.

Step 3: Read the summary

The AI summary is meant to be a starting point, not the final answer. It reads the same data you can see yourself — it just synthesizes it faster.

If the summary mentions a specific pattern, click through to the Patterns page to see the hourly chart and raw events.

Step 4: Review persistent patterns

Scroll down to the Persistent Patterns section. This shows patterns detected by the background cron job (runs every 5 minutes) and stored in the pattern registry.

Patterns marked New appeared for the first time since the last check. These are your most important signal — a new error pattern often correlates with a recent deploy or infrastructure change.

Step 5: Configure AI Digest (optional)

Click Configure Digest to set up automated AI summaries on a schedule.

Options:

The digest fires when the system detects elevated error activity. You won’t receive a digest for quiet periods.

Tip: Point the digest at a Slack channel dedicated to monitoring. You’ll get a plain-English summary of any error spikes without needing to open the dashboard.

You’re done

You now know how to:

AI summaries are generated using the Workers AI binding. The digest config endpoint allows configuring delivery programmatically.

← All guides