Insights
Is there an open standard for AI agent spend data?
Yes. The Open Agent Spend Attribution specification (OASA) is an open, CC BY 4.0 record format for attributing AI-agent spend to agent identity, task, and cost object, and for joining runtime telemetry, normalized billing, and payment settlement into a single auditable ledger row. It is built to interoperate with the two open standards that already cover the edges: OpenTelemetry's GenAI semantic conventions for what agents did, and the FinOps Foundation's FOCUS specification for what providers billed. Neither of those answers who the spend belongs to; OASA is the join.
Why the existing standards leave a gap
| Standard | Covers | Does not cover |
|---|---|---|
| OpenTelemetry GenAI semantic conventions | Model calls, tokens, latency, tool invocations as traces | Price, invoice, owner, cost object |
| FOCUS (FinOps Open Cost and Usage Specification) | Normalized billing rows across cloud and SaaS providers | Which agent or task generated the charge; unbilled private inference; settlement outside an invoice |
| Payment rails (card networks, x402 and similar) | That money moved, when, to whom | What it was for, in accounting terms |
Each of these is necessary. Put together they still do not produce the record a controller needs: this agent, on this task, for this cost object, consumed this, which was billed as that, and settled thus.
What OASA defines
One record, one event class, with a required envelope and optional groups so partial records stay valid and can be reconciled later:
- Envelope (required): record ID, record type (usage, charge, settlement, allocation, outcome), timestamps, source system, source record ID, currency.
- Agent identity: which agent, which version, which principal it acted for.
- Task and context: the workflow or task, trace ID linking to OpenTelemetry.
- Cost object: department, project, customer, GL account, the fields finance allocates by.
- Billing join: invoice ID and line reference linking to a FOCUS row.
- Settlement join: payment reference linking to the rail that moved money.
- Outcome (optional): the business result, for cost-per-outcome reporting.
The schema is additive-only within a major version, hub-and-spoke by design (canonical middle, adapters at the edges), and settlement-agnostic: an invoice, a card charge, a wire, and an on-chain payment reconcile into the same spine.
How it is used
- A gateway, agent runtime, or collector emits OASA usage records with agent identity and cost object attached.
- Billing adapters emit charge records from FOCUS-normalized invoices; settlement adapters emit settlement records from payment data.
- Allocation records express how shared usage was split, and outcome records attach results.
- The joined ledger produces accruals, chargeback statements, journal entries, and an audit trail from any GL line to the usage behind it.
Status and how to contribute
OASA is at version 0.1.1, published September 2026, maintained in the open on GitHub under CC BY 4.0. The full field reference is at onaro.io/spec. A mapping from OASA fields to FOCUS columns is in progress so that a FOCUS export from any FinOps platform can be joined to attribution data without transformation. Issues and pull requests are welcome; adapters for new gateways and billing sources are the most useful contributions.
Related
Not sure which of your AI costs are being booked? Run the free Agent Spend Assessment.
Onaro Meridian is FinOps for agentic AI: the system of record that attributes, controls and books what AI agents spend.
Brian Diamond
Brian Diamond is a fractional Chief AI Officer and founder of Onaro. He has spent 30 years running infrastructure operations and founded LANStatus, a Connecticut managed services provider and Microsoft partner, in 2001. He holds a Chief AI Officer certification and writes the CAIO Brief on AI leadership for finance and operations.