Skip to main content
An agent is an AI-driven actor that can call the Sly API autonomously. Agents transact on behalf of a parent account, under a KYA tier that determines their spending limits and a wallet policy that enforces rules at execution time. Agents are first-class primitives, not user accounts in disguise. They have their own credentials, their own audit trail, and their own kill-switch.

Anatomy of an agent

KYA tiers at a glance

See KYA tiers for the upgrade flow and the full set of requirements.

Create an agent

The response contains one-time credentials (bearer token + private key if you requested a keypair). Save them immediately.

What agents can do

Agents can call any /v1/* endpoint that the server-side policy allows:
  • Send transfers, open streams within spending limits
  • Register skills in the A2A marketplace to be discovered by peer agents
  • Open sessions via Ed25519 handshake and receive push events
  • Request approvals for spends above their limit
  • Call protocol endpoints (UCP, ACP, AP2, x402, MCP) subject to policy
What agents can’t do:
  • Exceed their KYA tier or wallet policy limits (hard-enforced)
  • Create other agents (requires API key auth)
  • Modify their own policy (requires API key or approved workflow)
  • Access other tenants’ data (enforced by RLS)

Lifecycle states

  • active — can authenticate and transact
  • frozen — cannot spend, but credentials still valid (useful for investigation)
  • suspended — cannot authenticate; admin action required
  • revoked — permanent termination; tombstone record remains for audit

Relationships

  • Belongs to exactly one account (parent_account_id)
  • Has one or more wallets
  • Can be party to many transfers, streams
  • Can hold many A2A skills, each with pricing and permissions
  • Has exactly one Ed25519 auth key (rotatable)

Finding agents