Understand Glacis
Glacis helps you apply controls to AI requests and responses, notice changes in their behavior, and keep signed records of what happened.
Your application sends supported model calls through the Proxy. Your organization’s Charter defines the configured controls and monitoring. A receipt records an evaluated operation and its results.
Start with your question
Section titled “Start with your question”| Question | Read |
|---|---|
| Where does Glacis run, and who operates it? | Deployment options |
| What can I configure? | Charters and controls |
| How do I investigate changes over time? | Drift monitoring |
| What does signed evidence establish? | Receipts and trust |
| How can I check a downloaded receipt? | Verify a receipt |
| What should I confirm before sending sensitive data? | Privacy and responsibilities |
What coverage means
Section titled “What coverage means”Glacis covers the supported operations routed through your configured deployment. Calls that bypass it are outside that coverage. Authentication or connection failures can happen before a receipt exists.
A control result reports what that control evaluated. It does not guarantee that a model’s answer is correct or that every risk was detected. Your team remains responsible for deciding which controls and evidence meet its needs.
Use your deployment
Section titled “Use your deployment”Sign in to the Glacis portal to see your organizations, deployment status and available controls. Use your deployment’s connection information rather than copying another organization’s endpoint or credentials.
Detailed setup and operating instructions are in the private customer handbook. Ask your Glacis contact for access.