ControlTheory Launches Dstl8 2.0 AI SRE With Metrics, Traces and GPU Visibility

ControlTheory unveiled Dstl8 2.0, an advanced AI site reliability engineering platform, at KubeCon + CloudNativeCon North America 2026. This latest iteration builds upon ControlTheory's Telemetry Distillation methodology, expanding its capabilities to incorporate metrics, traces, and GPU visibility. Unlike previous versions, Dstl8 2.0 autonomously collects telemetry data, eliminating the need for additional observability platforms.

The platform meticulously tracks AI workloads from the initial agent interaction through model calls, encompassing routing through gateways and concluding at GPU performance. ControlTheory's CEO, Bob Quillin, emphasized the platform's ability to diagnose and address issues solely based on its internal telemetry, questioning the necessity of traditional observability tools.

Dstl8 2.0 leverages traces to identify performance issues like slow endpoints, error rate spikes, and novel request patterns, linking these events to corresponding logs, metrics, and associated deployments. Upcoming metrics feature predictive analytics for disk, CPU, and GPU capacity thresholds, with the platform harmonizing all signals with Kubernetes events such as deployments, restarts, and memory exhaustion.

Additional enhancements include customizable dashboards, compatibility with Prometheus and Jaeger, and a streamlined setup process for Kubernetes, AWS, and OpenTelemetry sources. Dstl8 2.0 is now accessible, with metrics and traces currently in preview.