Revenium unveils Tool Registry to track full AI costs

Revenium has launched Tool Registry, a product that tracks the full cost of running AI agents across external services, internal tools and human review, rather than focusing only on large language model token usage.

Revenium positions the product as a way for finance and technology teams to connect agent activity with spend that otherwise shows up only as line items in vendor invoices. It also includes controls that can stop an agent workflow when it reaches a predefined cost ceiling.

Many organisations are deploying so-called agentic AI in business processes. These systems often combine an LLM with software tools and data sources. They may call third-party APIs, query paid datasets, use SaaS applications or route tasks to staff for review. Each step can incur a fee that does not appear in the model bill.

Revenium argues that token monitoring provides only a partial view of spend, with larger costs often coming from per-transaction charges from external providers and labour for review steps.

A loan origination example illustrates the gap. Revenium estimates that LLM tokens might cost $0.30 for a single workflow, while the same workflow could pull a credit report for $35 to $75, run identity verification for $2 to $5, check fraud scores for $1 to $3, and verify bank accounts for $0.25 to $1. Total cost per application can reach $50 to $85, with tokens accounting for less than 1%.

Those charges typically appear in monthly invoices from vendors such as Equifax, Westlaw or Qualtrics. They do not usually map back to the workflow, decision or customer interaction that triggered them, making it difficult to understand the unit economics of an agent-driven process.

John Rowell, Chief Executive Officer and Co-founder of Revenium, said the product is meant to answer questions about return on investment.

"Almost every CFO eventually asks the same question: is AI actually saving us money? And no one can give a clean answer," Rowell said.

Rowell said cost and outcome data often sit in separate systems and invoices.

"The costs live across a dozen vendor invoices. The outcomes live in a CRM somewhere else. Tool Registry closes that loop. Every dollar an agent spends is attributed back to the exact decision that triggered it, in one place," he said.

Cost attribution

Tool Registry records cost sources and attributes them to specific execution details. Revenium says organisations can register external REST APIs, MCP servers, SaaS platforms, internal compute functions and human review time. It then meters each invocation back to the agent, workflow, trace and customer associated with the call.

The product includes a pricing-model layer that can handle per-call charges and tiered pricing. Revenium also describes an auto-discovery feature for tools detected at runtime that have not been registered.

It also includes circuit breakers that halt execution when per-trace or per-workflow cost ceilings are reached, shifting cost control from after-the-fact reporting to enforcement during execution.

A unified analytics view displays token costs alongside tool costs in a single dashboard and can break spend down by organisational unit, product, agent and customer.

Human review

Revenium also treats human-in-the-loop activity as a tracked cost event within the same trace as tool calls. This structure makes human review time another metered component of an agent workflow, alongside model inference and external services.

Revenium highlights the use of such data in automation programmes where the share of cases requiring human review falls over time. It cited an insurance claims workflow that starts at 35% human review and drops to 12% over six months. Revenium says it can measure both the changing labour component and machine-driven costs within the same record.

Jason Cumberland, Chief Product Officer and Co-founder of Revenium, said the company sees external spend as the core issue in AI economics.

"The real AI cost problem is not token pricing, it is agent-driven spend happening outside the LLM bill. Every workflow can trigger multiple external services with separate invoices and no attribution. Tool Registry gives organizations the first complete view of agent-initiated spend across their entire stack," Cumberland said.

Market backdrop

Industry forecasts point to rising adoption of task-specific AI agents in enterprise software. Gartner projects that 40% of enterprise applications will feature task-specific AI agents by the end of this year, up from less than 5% in 2025. Forrester's 2026 Technology and Security Predictions report projects enterprises will defer 25% of planned AI spend into 2027 due to ROI concerns, linking the trend to difficulty measuring whether a workflow produces more value than it consumes.

Against that backdrop, vendors are broadening cost management from model monitoring to workflow-level governance, including spend analysis across multiple services and controls that operate at runtime rather than in monthly reporting cycles.

Tool Registry is available under a free developer tier, SMB pricing and custom enterprise plans. Revenium said it expects organisations to use the product as agent deployments expand across more business processes and incorporate a wider range of paid tools and review steps.