Matcha by Keeya Labs addresses the challenge of understanding and managing energy consumption in AI workloads by connecting GPU power telemetry with detailed workload traces. The platform identifies which models, training runs, inference requests, tenants, and workflows are responsible for energy use, offering organizations insight into their AI infrastructure’s power consumption patterns.
The tool reads GPU telemetry, workload traces, and cluster metadata to explain energy spikes, highlight inefficient workloads, and suggest areas for optimization. Notably, Matcha provides workload-level attribution, showing GPU energy usage broken down by model, training run, inference request, tenant, and workflow. It offers a unified dashboard for monitoring key performance indicators, trends, and activity feeds, giving users a central view of their AI operations. The platform includes efficiency diagnostics and delivers alerts on issues such as idle GPU waste, helping teams detect and address inefficiencies in real time. Cost signals and estimated costs by workload are also surfaced, enabling organizations to compare training runs, inference services, and agent workflows by energy, cost, and efficiency.
A conversational AI agent is integrated into Matcha, allowing users to ask questions about changes in energy usage, identify where energy went, and determine which workloads need attention. The tool is designed to integrate with existing AI infrastructure stacks, plugging into telemetry, workload, and cluster tools already in use, so teams do not need to migrate their systems.
Matcha is aimed at AI infrastructure teams, including those operating GPU clouds, AI labs, or enterprise clusters. It supports organizations that run training or inference workloads and need to bring accountability to shared AI clusters by providing visibility into energy and cost by team, tenant, model, and workflow. The platform is positioned as a solution for tracking customer, model, and workload-driven power use across fleets, supporting both operational efficiency and cost management in AI environments.
In the LLM eval & observability space, Matcha by Keeya Labs takes a focused approach. It focuses on understanding and optimizing the energy consumption of AI workloads by correlating GPU usage with specific models and workflows. It is built as a B2B product for AI infrastructure teams and data engineers. The product ships for the web and the command line.
It is developed by Keeya Labs, and the product first shipped in 2026. Development happens publicly on GitHub with 33 commits in the last 90 days. Key capabilities include GPU telemetry, energy usage analytics, and workload tracing.
Latest indexed changes and source events
Other apps tracked under the same category.
