Grafana Launches AI-Powered Assistant to Slash Database Performance Troubleshooting Time

Breaking: Grafana Cloud Introduces AI Assistant for Database Observability

Grafana Cloud has released a new AI-powered assistant integrated directly into its Database Observability platform, enabling engineers to diagnose slow queries and performance bottlenecks in seconds rather than hours. The assistant analyzes live Prometheus and Loki data, real table schemas, and execution plans—eliminating the need to manually gather context.

“This isn’t another generic chatbot. The assistant runs against your actual database environment, with your schemas and time windows preloaded,” said Sarah Chen, Senior Product Manager for Database Observability at Grafana Labs. “Engineers can now move from seeing a latency spike to understanding its root cause in a single click.”

What the Assistant Does

The assistant provides purpose-built analysis actions designed by database engineers, not generic prompts. Users click predefined buttons—such as “Why is this query slow?”—and the assistant synthesizes data from Prometheus (metrics) and Loki (logs) to produce a health assessment.

For example, it can detect that a query’s P99 latency is 12 times its median, indicating an intermittent problem, or that wait events consume 40% of execution time—even translating cryptic names like wait/synch/mutex/innodb into actionable advice.

Background

Traditional database troubleshooting forces engineers to manually extract SQL, copy it into separate AI tools, and reconstruct schema and time ranges. Grafana Cloud Database Observability already offered RED metrics, execution samples, wait event breakdowns, table schemas, and visual explain plans—but visibility alone didn’t speed up diagnosis.

The new integration closes that gap by embedding AI directly into the workflow. No data is stored or used for model training; query text and schema metadata are used only for the current analysis session.

What This Means

For DevOps and SRE teams, this translates to faster mean time to resolution (MTTR) for database performance incidents. Instead of spending 30 minutes correlating metrics and logs, engineers get a synthesized answer in under a minute.

“We designed it so that any engineer, regardless of database expertise, can interpret complex wait events and execution plans,” noted Chen. “It’s like having a senior DBA on call 24/7.”

Key Features at a Glance

• Contextual AI analysis — Uses your actual Prometheus and Loki data sources in the time window you’re viewing.
• One-click prompts — Pre-defined analysis actions for slow queries, degraded performance, and optimization recommendations.
• No data leakage — Query text and schemas are ephemeral; never stored or used for training.
• Human-readable explanations — Translates obscure wait event names into plain English with specific advice.

How It Works in Practice

When a query shows a P99 latency spike, an engineer clicks the assistant button. The tool queries both Prometheus and Loki for the selected time range, then synthesizes a single health assessment. For instance, it might reveal that “rows examined are 50 times rows returned,” meaning wasted filtering work, while CPU remains healthy but wait events dominate.

The assistant also identifies intermittent vs. constant issues by comparing P99 to median latency. It can pinpoint lock contention, bad joins, or table scans that only become problematic as data grows.

Availability

The Grafana Assistant integration for Database Observability is available now for all Grafana Cloud customers. No additional configuration is required—it activates automatically in the query detail view.

For a deep dive, see the Background section above or read the official documentation.