Grafana Labs adds AI Watchers and code-fixing agents to Grafana Cloud
Grafana Labs announced a wave of Grafana Cloud AI features at ObservabilityCON 2026 on October 20, headlined by generally available Assistant Watchers and Agent Observability with OpenTelemetry support, plus a public preview of Grafana Managed Sandboxes, which can prepare code fixes and open pull requests for human review.
Assistant Watchers lets teams describe what to monitor in plain language. A Watcher learns patterns from telemetry and surfaces evidence when those patterns change, routing findings through existing Grafana Alerting; a critical finding can kick off a deeper Assistant Investigation, a capability GA since July.
Grafana Managed Sandboxes, in public preview, gives the built-in Assistant a controlled cloud environment to examine repository code alongside live telemetry and open a pull request with a proposed fix. The preview deliberately stops at a reviewable change - a human still approves anything that merges.
For teams running AI in production, Agent Observability's GA OpenTelemetry support brings agent activity - prompts, tool calls, responses, latency, token consumption, model costs and evaluations - into the same instrumentation standard as other services, helping separate an expensive model call from a slow downstream service.
Grafana is also placing a Knowledge Graph in every Grafana Cloud stack, linking applications, databases, Kubernetes infrastructure and user-facing telemetry. Custom entities and runtime context for agents are GA, and a new Cypher query capability lets agents pull relationships from the graph via gcx.
On costs, Grafana Cloud Logs gets GA full-text search enhancements from Logline (acquired April 2026) at no extra charge, Adaptive Metrics Auto-Apply is GA with a Cost Mode to cap active series growth, and Adaptive Logs Archive enters public preview, routing logs to customer-owned object storage.