DoiT launches Attribute™ to trace AI token and GPU costs at the kernel level

As the cost of running AI agents and large language models scales across enterprise cloud environments, the demand for granular consumption data has outpaced what most observability tools were built to provide. DoiT, the Santa Clara company behind Cloud Intelligence™, introduced Attribute™ on July 7, a product it says measures real-time AI consumption at the kernel and connects every token, model request and GPU cycle to the specific customer, feature or agent that drove it. The tool is designed to work without tags, SDKs or code changes.

What Attribute™ measures and how

The kernel-level approach is the distinguishing technical claim. Rather than relying on developer-added instrumentation, Attribute™ intercepts consumption data where the compute actually happens. The product maps every token, model request and GPU cycle to its origin: a customer account, a product feature or an autonomous agent running in the environment.

The no-code-change requirement has practical weight. Engineering teams building on AI infrastructure have typically faced a choice between manual tagging, which is error-prone and inconsistent across teams, and purpose-built SDKs that add integration work and dependency overhead. Attribute™ operates below the application layer and sidesteps both.

Sector context: AI cost visibility as a distinct infrastructure problem

The launch sits inside a broader cycle in cloud and AI infrastructure. As AI agents proliferate and inference workloads grow, the gap between what enterprises spend on compute and what finance teams can attribute to a product line or a client account has widened. Cost visibility has become a distinct product category.

DoiT's positioning of Attribute™ alongside Cloud Intelligence™ suggests the company sees AI tokenomics, the economics of token consumption and GPU allocation, as infrastructure to be managed at the kernel rather than the application layer.

The macro caveat is adoption friction. Kernel-level instrumentation requires access and vendor trust that enterprise security teams may scrutinize carefully before granting.