Prompt-based
Prompt-based Debugging allows users to initiate investigations by describing issues in natural language. This feature is useful when you’re unsure where to start or when you want to quickly delegate an exploratory analysis to the AI agent.
Instead of needing to navigate through dashboards or manually gather logs, you simply type what you’re observing — and the agent uses its reasoning engine and available integrations to begin debugging.
Agent Response:
🧠 How It Works
- Describe the issue in a prompt — as you would to a teammate.
- The agent interprets the intent, identifies relevant data sources and tools, and begins fetching context.
- It analyzes logs, metrics, configurations, or infra data to identify potential causes.
- Based on findings, it surfaces:
- Observations from the environment
- A likely root cause (if any)
- Suggested remediation actions
📘 Example
Prompt:🔎 Investigation Summary
- Queried CPU metrics from Datadog for checkout-service pods.
- Found average CPU usage at 92% over the last 15 minutes.
- Analyzed top CPU-consuming processes via top command on the related nodes.
💡 Insight:
- checkout-worker pod running Celery processes is using 55–75% CPU on each replica.
- Spikes correlate with batch job executions at :00 and :30 marks.
📌 Root Cause:
- Worker pods overwhelmed during scheduled batch jobs due to insufficient CPU requests.
🛠️ Suggested Actions:
- Increase CPU requests/limits for checkout-worker pods.
- Consider spreading batch jobs or introducing queuing.
🧰 When to Use Prompt-based Debugging
- When no alert is raised but something “feels off”
- During exploratory analysis
- For recurring issues where you want a fresh look
- When you want a quick sanity check from the AI agent
🔐 Notes
- Prompts work best when specific — include service name, time window, or symptoms.
- The agent will fall back to general guidance if no integrations are configured.