Concept · L2

Inference — L2.

A single model call routed through Ainfera — signed by the Agent Card, settled from a Wallet, and recorded as AuditEvents.

[02]What it isOne-paragraph definition

An Inference is the core verb of the platform. One API call resolves to a Provider and Model under the Tenant's spend policy, debits the Wallet, and emits a chain of AuditEvents. It is the atomic unit that everything else is recorded against.

[03]Where it shows upAPI · SDK · Audit log

Where it shows up.

API: POST /v1/inferences · GET /v1/inferences/:id
SDK: from ainfera import Inference
Audit log: appears as inference.requested, route.selected, provider.responded events

[04]Code exampleWorking snippet

Code example.

out = client.inferences.create(
    agent_card=card,
    model="claude-opus-4-7",
    messages=[{"role": "user", "content": "hello"}],
)

print(out.text)
print(out.provider)        # which Provider served it
print(out.receipt.url)     # signed Receipt
print(out.audit_chain_id)  # hash-chained AuditEvents

[05]RelatedOther concepts · full ontology

Spec alignment

Routing behavior follows ATS v1.0; cost reporting is structured for Annex IV-compatible technical documentation.