A unified portal to assess compliance, explore controls, and deploy your AI Gateway.
These are illustrative sample figures. Connect your account in Step 3 and a live-data switch appears right here.
DLP policy logs show 3,687 attempts to upload credentials, source code, financial figures, and PII to external models were blocked. Hitting any unmonitored endpoint risks significant compliance liability.
Guardrails scanned all prompts, blocking 1,182 requests targeting jailbreaks (558), PII (376), hate speech (133), and violence (97) before they reached the model.
Only 2 of our 7 sanctioned AI applications have all security controls active. We've identified 2 High-risk applications (including our HR resume screener) running with Logging, Auth, and DLP completely disabled.
| Application | Risk | DLP | Guardrails | Auth | Logging |
|---|---|---|---|---|---|
| northwind-support-copilot | Limited | ✓ | ✓ | ✓ | ✓ |
| checkout-rag-assistant | Limited | ✓ | ✓ | ✓ | ✓ |
| hr-resume-screener | High | ✗ | ✓ | ✗ | ✗ |
Our support copilot served 37.7% of its total requests directly from the edge cache, avoiding $890 in billing. One team (Returns) has reached 95% of its monthly budget, triggering a predictive threshold alert.
Using Dynamic Routing, we split traffic based on user context (e.g. sending C-Level users to Claude 3.5 Sonnet and routing standard queries to Workers AI). If the primary provider experiences an outage, AI Gateway automatically fails over to an active fallback, preserving 100% uptime for end-user applications.
By leveraging the evaluations API (patchLog()), our applications capture thumbs up/down feedback and model-as-a-judge score ratings (0–100) directly on our gateway logs. This helps us track prompt drift, monitor response degradation, and build evidence for model selection decisions.
Every tool called by an autonomous AI agent is logged in our Model Context Protocol (MCP) Server Portal, attributing activity to specific user email identities. Failed calls, schema syncs, and upstream latencies are fully logged, and outbound agentic requests are scanned by Cloudflare Gateway and DLP to prevent data exfiltration via third-party tools.
Step through the actual Cloudflare AI Gateway interface. See how logs, analytics, guardrails, and dynamic routing work in a structured, guided walkthrough over realistic snapshots.
Deploy a live AI Gateway (ai-control-plane-demo) directly on your account. Includes pre-configured budgets, caching, and safety rules.
AI Gateway: Edit · Workers AI: Edit · Zone: Read · Account Settings: Read · Expires: 24h
Prefer not to paste a key into someone else's site? This portal is fully open source. Run it yourself from github.com/kai-cloudflare/ai-control-plane-demo — run it locally, deploy it as a Worker on your own account, or spin up a fresh preview account, all from the repo. When you run it locally your token never leaves your machine.