Cost attribution
How Silvanexum rolls agent spend up the multi-hop agent→agent→tool mesh, so every dollar traces back to the agent that caused it.
Every run is metered and signed; every agent-to-agent call is recorded on the hop ledger. Together they answer the question a provider dashboard can't: not "how many tokens did I use," but "which agent — on whose behalf — spent this, and was it authorized?"
Per-run metering
When a run completes, Silvanexum emits a metering event priced from the signed trace: tokens in/out, model, latency, and the credits the run consumed. Because it is derived from the signed record, the cost is reproducible and tamper-evident — not a number you take on faith.
Multi-hop, across the mesh
When one agent calls another (A2A), the call is recorded on the ledger — escrowed on entry, settled on completion — and attributed to the caller, the callee, and the grant it ran under. Cost rolls up the chain, so a charge deep in a sub-agent still traces back to the top-level agent (and customer) that triggered it. This multi-hop attribution is the part a proxy or single-call meter cannot do.
Attribute by any dimension
The same metered events roll up by agent, project, model, or provider
(GET /spending/summary?groupBy=…), so you can answer "which agent ate the
budget this week" or "which model/provider is the spend concentrated in" — not
just an aggregate bill.
Attribution is the foundation; acting on it — budgets, policies, and blocking runaway or unauthorized spend before it bills — is covered in spend control.