Verifiable Agent Audit and Cross-Framework Observability

The problem

• Agent activity and policy decisions are hard to tie together across frameworks.
• Ordinary logs miss the calls that were denied.

What MACAW does

• One import adds policy, signed audit and OpenTelemetry to OpenAI and Pydantic AI apps.
• Four calls, two of them denied, appear in the MACAW console.
• The same calls stream to Grafana, with no logging code.

What’s different

• Each decision comes with its evidence.
• Signed audit records beside operational telemetry.
• One policy and evidence model across frameworks.

Policy and deployment

Policy

Deny GPT-4 and dataset exports. Record every call.

Deploy

The SecureAI adapters or the macawClient SDK in your app, with policy in the MACAW control plane.