Runtime Certification
Certify your agent against contracts
Overview
Agent Contracts can be used in runtime to certify if your agent’s behavior matches the defined contracts.
How certification works
Certification in AI systems refers to the process of attaching some form of proof, or certificates, to system outputs that verify their correctness according to predefined contracts. A certificate serves as a guarantee that the output satisfies all specified requirements - not just in terms of the final result, but also in how that result was obtained.
This certification process transforms what would otherwise be a “black box” AI output into a verifiable result with a clear chain of evidence demonstrating its correctness. The certificate itself contains the proof that each contract condition was satisfied, and the evidence that the system followed the correct process, allowing any party to independently verify the system’s behavior.
The power of certification lies in its ability to provide objective, measurable guarantees about system behavior. Rather than simply trusting that an AI system “probably” did the right thing, certification gives us concrete evidence that it definitively followed all required steps and produced a valid result according to our specifications.
Certify your agent execution
Prerequisites:
- Installed and set up agent contracts for runtime certification installation.
- Created a runtime specification for your agent contract specifications.
Start your certification workers
Start the certification service with the specs.json
file and the runtime-certification.yaml
configuration file.
Retrieve certificates in real-time
How to use certificates
Certificates are can be retrieved in real-time and be used to inform your application of how well the agent is following the contracts. You can use the information to take custom actions (e.g. escalate to human, send a notification, retry, etc.)