Skip to content

Control

Control is the operational boundary layer for ThinkWork. It answers two blunt questions:

  • what are agents allowed to do
  • what actually happened when they did it

That means Control is not just safety policy. It is also spend management, visibility, and auditability.

In the current open source docs, Control includes:

  • guardrails at model invocation time
  • budgets and usage tracking
  • rate limits and spend boundaries
  • audit logging in your AWS account

As soon as agents move from demos into real work, you need a way to enforce boundaries and inspect behavior after the fact. ThinkWork treats that as a first-class concept instead of an afterthought.

Guardrails

Bedrock guardrails and the enforcement boundary for managed agents.

Budgets, Usage, and Audit

Spend controls, turn-level usage tracking, and durable audit logs.