Skip to main content

Introduction to Caracal Core

Caracal Core is a network-enforced policy enforcement and metering engine designed specifically for AI agents. It sits between your agents and the external APIs they consume, ensuring that every request is authorized, metered, and auditable.

Why Caracal Core?​

As AI agents become more autonomous, they need robust controls to prevent:

  • Runaway Spending: Agents making unlimited API calls.
  • Policy Violations: Agents accessing unauthorized resources.
  • Unaccountable Actions: Lack of audit trail for agent behavior.

Caracal Core solves these problems by enforcing policies at the network level, not relying on agent self-reporting.

Core Concepts​

Gateway Proxy​

All agent traffic flows through the Caracal Gateway, which intercepts requests, evaluates policies, and meters usage in real-time.

Policy Engine​

Define spending limits, time windows (daily, weekly, monthly), and resource allowlists. Policies are evaluated before every request.

Immutable Ledger​

Every metering event is recorded in a Merkle tree-backed ledger, providing cryptographic proof of all spending.

Fail-Closed Design​

If the policy engine or ledger is unavailable, requests are denied by default, preventing unchecked spending.

Next Steps​