Operator Tiers

Five tiers of access, from read-only to kill switch.

Every operator and every agent holds one of five tiers. A tier decides which actions are permitted, and it is enforced on the server, so it cannot be escalated through a request or a prompt. Each tier inherits everything the tier below it can do.

The controls described here are verified by open benchmarks anyone can run: every one of GovernanceBench’s 235 real-world governance scenarios passes, and all 148 of VIGIL’s documented adversarial attacks are defended. See the benchmarks.

Tier 1 Read only, enhanced monitoring
Evaluator
For auditors, regulators, and anyone verifying the claims before they trust the system.
  • Read the full audit trail and confirm every action was recorded and left unaltered.
  • Run the open benchmarks against any deployment and check the scores yourself.
  • Verify the signed identity of any agent in the fleet.
  • Change nothing: no action at this tier can alter a running system.
Next Analyst adds standard read access for everyday research and review.
Tier 2 Read only
Analyst
For a developer starting out, or anyone adding governance to an agent that already exists.
  • Browse the fleet, the dashboards, and the reports to see how agents actually behave.
  • Review each governance decision and the reasoning recorded behind it.
  • Send events to your own security monitoring and logging tools.
Next Builder adds write access, so you can wire governance into the agents you build.
Tier 3 Read and write
Builder
The day-to-day working tier for development and integration.
  • Wrap an existing agent so every action it takes is governed and recorded.
  • Issue and verify a signed identity for the agents you build.
  • Connect those agents to your secrets manager, so credentials are never hard-coded.
  • Record a tamper-evident history of every workflow your agents run.
Next Operator adds the controls to run a fleet in production: halt, resume, and quarantine.
Tier 4 Execute, deploy, configure
Operator
For the team running an agent fleet in production and answering for it when something goes wrong.
  • Deploy and configure agents across a production fleet.
  • Halt a single agent instantly, resume it, or quarantine it to a limited set of permissions.
  • Watch fleet-wide health and export the audit trail to a regulator on request.
  • Cannot switch governance off: the rules that constrain the agents still bind you.
Next Strategist adds the top-level controls: halt the entire fleet, or any scoped subset, at once, and decide who holds which tier.
Tier 5 Full access
Strategist
The kill-switch tier, for the executive or risk owner accountable for the whole fleet.
  • Halt the entire fleet in one command, then resume it once the incident is over.
  • Assign or revoke the tier held by any operator or any agent.
  • Set the governance rules that every other tier operates under.
  • Every Strategist action is written to the same tamper-evident trail as everyone else.
Next Strategist is the top tier: nothing sits above it.

Pick the lowest tier that covers what you need, and upgrade when the next capability is required.