Revenium has announced the general availability of its Tool Registry, a new capability designed to give enterprises a complete, end-to-end view of what their AI agents actually cost. Moving beyond traditional token-based tracking, the platform attributes every expense, across APIs, third-party services, and even human intervention, back to the specific AI decision that triggered it, providing what the company claims is the first full-stack view of agent-driven spend.
While most organizations closely monitor large language model (LLM) token usage, Revenium argues that tokens represent only a fraction of the true cost of running AI agents. In real-world workflows, agents routinely call external services, such as credit bureaus, identity verification platforms, fraud detection tools, and data providers, each incurring transaction-based fees that are often significantly higher than the cost of the model itself.
A typical example illustrates the imbalance. In a loan origination workflow, token usage might cost as little as $0.30. However, the same workflow could trigger a credit report costing between $35 and $75, identity verification at $2 to $5, fraud checks at $1 to $3, and bank account verification at up to $1. In total, the workflow may cost between $50 and $85, with token usage accounting for less than 1% of the total spend. These external costs are usually buried in separate vendor invoices from providers such as credit agencies, legal data platforms, or survey tools, making it nearly impossible to trace spending back to specific AI-driven actions.
Revenium's Tool Registry is designed to close that visibility gap. The platform allows organizations to register and track any cost source involved in an AI workflow, including external APIs, SaaS platforms, internal compute services, and human review steps. Each invocation is then mapped back to the originating agent, workflow, transaction trace, and even the end customer involved. This creates a unified system of record in which every dollar spent by an AI agent can be directly tied to a business outcome.
According to John Rowell, CEO and co-founder of Revenium, the lack of cost transparency has become a recurring challenge for executives evaluating AI investments. He notes that organizations often struggle to determine whether AI is delivering real financial value because costs are fragmented across multiple systems, while outcomes are tracked elsewhere. By linking spend directly to agent decisions, the Tool Registry aims to provide a clear answer to whether AI workflows are generating a return on investment.
The timing of the release reflects broader industry trends. Gartner projects that 40 percent of enterprise applications will include task-specific AI agents by the end of 2026, a sharp increase from less than five percent in 2025. At the same time, Forrester has predicted that organizations will defer a quarter of planned AI spending into 2027 due to uncertainty around ROI. A key driver of this hesitation is the inability to measure whether AI-driven workflows are actually producing more value than they consume.
A notable aspect of the platform is its inclusion of human-in-the-loop costs. In many enterprise workflows, AI agents still rely on human review for validation, compliance, or exception handling. The Tool Registry treats these human interventions as measurable cost events within the same execution trace, enabling organizations to quantify how automation impacts human effort over time. For example, a workflow that reduces human review from 35 percent to 12 percent over several months can demonstrate tangible cost savings, but only if both human and machine contributions are tracked within the same system.
Most existing tools that measure AI cost today take a model-centric or observability-first approach, which is fundamentally different from Revenium's approach.
Platforms such as Langfuse, LangSmith, and Helicone focus primarily on tracking LLM usage, tracing agent workflows, and monitoring token consumption. They provide detailed visibility into prompts, responses, latency, and per-request costs, often breaking this down by user, feature, or session. Many of these tools use proxy- or SDK-based approaches to capture telemetry and can provide cost dashboards, alerts, and evaluation frameworks to assess output quality. However, their definition of "cost" remains largely tied to LLM interactions and infrastructure-level metrics, with limited visibility into what happens beyond the model call.
Even more advanced platforms, such as Arize AI and Datadog LLM Observability, extend this by combining performance monitoring, tracing, and evaluation, giving enterprises a unified view of model behavior and system health. These tools can trace multi-step agent workflows and provide end-to-end telemetry. However, they still tend to treat cost as a byproduct of system usage (tokens, compute, requests) rather than a fully attributed business expense tied to external services and outcomes.
Revenium's approach diverges by shifting from "AI observability" to "AI financial attribution." Instead of asking how much a model call costs, it asks what an entire agent decision costs in real terms, including external APIs, SaaS platforms, and human intervention. Most current tools do not natively capture this layer. While some observability platforms can trace tool usage within an agent workflow, they typically do not normalize and attribute third-party billing (e.g., per-API transaction fees or human review costs) back to a single decision or customer-level outcome.