How Agents Use Dino

User-facing workflow for governed agent purchasing in Dino.

Dino is the governance and audit layer for agent purchasing.

The practical workflow is:

agent request -> policy/approval decision -> scoped payment path -> auditable lifecycle

#Roles

#Agent

  • proposes what to buy
  • calls your constrained tool boundary
  • reports status back to the user

#Your tool/service

  • stores Dino credentials safely
  • injects auth + idempotency
  • calls Dino purchase-intent APIs

#Dino

  • enforces budgets, thresholds, merchant rules
  • records spend and lifecycle events
  • emits webhooks for async control loops

#End-user flow

  1. User asks the agent to buy an item.
  2. Agent creates a purchase intent in Dino.
  3. Agent authorizes the intent (policy + approval check).
  4. If approval is required, operator approves in dashboard.
  5. Agent/tool issues a scoped payment credential handle.
  6. Execution adapter performs checkout.
  7. User and operators track status through webhooks and status reads.

#Status mapping for product UI

  • intent_created -> request created
  • approved -> approved for execution
  • needs_approval -> waiting for human approval
  • declined -> declined by policy
  • payment_credential_issued -> payment path prepared
  • execution_started -> checkout in progress
  • completed -> purchase completed
  • failed -> purchase failed

#Security boundaries

Do:

  • keep spend keys and payment secrets server-side
  • use a constrained tool/API boundary for model calls
  • use idempotency for write operations

Do not:

  • put keys or card data in model prompts/transcripts
  • expose arbitrary HTTP request builders to the model