Executions

Every agent run produces:

One execution_log row
Zero or more execution_steps
Optional billing usage rows

Step format

{
  "stepIndex": 0,
  "type": "tool_call",
  "toolName": "check_balance",
  "toolInput": { "accountId": "..." },
  "toolOutput": { "balance": 24500 },
  "durationMs": 340
}

Execution error codes

AGENT_NOT_FOUND
AGENT_INACTIVE
MODEL_UNSUPPORTED
EXECUTION_FAILED
UNAUTHORIZED
APPROVAL_REQUIRED
TOKEN_BUDGET_EXCEEDED

These codes are defined in packages/ai/src/types.ts.