Skip to main content

Operator & Governance

The Operator is you — the human who runs the SLAW instance and retains full control over what the agents do.

SLAW is designed for autonomous AI work, but autonomy without accountability is a liability. The Operator layer exists so that a human always has the last word: on hires, on strategy, on spend, and on whether any agent keeps running.

The full hierarchy​

Instance  ←  one local SLAW deployment
└── Operator  ←  the human governance actor
    └── Squad  ←  the AI team
        └── Squad Lead  ←  the team's lead agent
            └── Agents

One instance runs many squads. One Operator governs all of them. The Operator is not a squad member — they sit above the squad boundary and have visibility and control across every squad on the instance.

What the Operator controls​

ActionWhat happens
Approve or reject hiresAgents can be gated — the Squad Lead requests a hire, the Operator approves or blocks it
Review and approve strategyBefore a Squad Lead breaks a goal into tasks, the Operator can review and approve the plan
Pause or terminate any agentPause stops an agent from receiving new heartbeats; terminate ends its involvement permanently
Override decisionsThe Operator can intervene in any active task — reassign, redirect, or cancel
View the audit logConsequential mutations are logged; the Operator can inspect the activity trail (actions and field changes, not run transcripts)
Set and enforce budgetsMonthly spend limits per agent and per squad; see Costs & Budgets

Operator API key​

The Operator authenticates with a key prefixed slaw_op_. This key is generated during instance setup and never leaves the machine. It gives full write access to the control plane — keep it out of agent configs and source control.

Approval gates​

Approval gates are the main mechanism for keeping humans in the loop without slowing everything down. You decide which actions require your sign-off:

  • Hire approval — the Squad Lead cannot create new agents until you approve each hire request. Useful when you want to review cost or capability before a team expands.
  • Strategy approval — the Squad Lead drafts a plan for a new goal and waits for your approval before breaking it into tasks. Ensures the direction is right before work begins.

Gates are configurable — turn them on for sensitive squads, off for trusted teams running on well-understood goals.

Pause and terminate​

Both operate on the org tree:

  • Pause — the agent stops receiving heartbeats. It holds its current state. Useful for reviewing an agent's recent work or freezing spend. Resumable.
  • Terminate — the agent is removed from the active roster. Its history is retained for audit purposes. Not resumable.

Because agents live in a tree, pausing a lead agent pauses the entire branch below it. This is useful when you want to halt a sub-team without touching the rest of the squad.

The audit log​

Every consequential action — task assignments, status changes, budget events, hire approvals, strategy decisions — is written to the activity log. The Operator can filter by agent, squad, or time range. The log is append-only and cannot be edited by agents.

Instance-level vs tower-level governance​

On a single SLAW instance, the Operator has full local control. When an instance is enrolled with a SLAW Botfather tower, governance extends upward: the tower can set fleet-wide budget limits and the tower admin manages enrollment and revocation. Local Operator controls remain in place — the tower adds a layer above, it doesn't replace the one below.

Next steps​