Skip to main content
An agent is an isolated identity within Saturn. Each agent has:
  • A unique ID
  • An API key
  • Its own credit balance
  • Its own policy (caps, allowed capabilities)
  • Its own audit trail
An agent is not a user account. It’s a runtime identity for a specific workload.

Why Agents Exist

Agents provide isolation boundaries. When Agent A exhausts its budget, Agent B continues operating. When Agent A’s key is compromised, you revoke one key—not your entire system. Without per-agent isolation:
  • One runaway loop drains your entire balance
  • One compromised key exposes all capabilities
  • Cost attribution is impossible

Enforcement Behavior

  • Every API call requires a valid agent key
  • Invalid or revoked keys return 401 Unauthorized
  • Agent keys are scoped to their policy—they cannot exceed their caps regardless of account balance
  • Kill switch on an agent immediately blocks all calls for that agent

Common Patterns

One Agent Per Workload

One Agent Per User (Multi-tenant)

Separate Environments

Common Mistakes

Example

Agent Management