Coming soon · design previewThis page mocks up an unreleased capability with realistic data. It is not yet wired to live providers. Back to demo home

Coming soon · AI Gateway + DLP

Inspect every AI API call. Redact PII. Enforce policy at the edge.

Cloudflare AI Gateway sits in front of every approved provider. Every call is logged with token counts, user attribution, redactions applied, and policy decision. Approved-models policy is enforced server-side — banned models get blocked before they leave your network.

Calls inspected · last 24h

12,847

Redactions applied

Policy blocks

Cost saved via smart routing

$214

Live call feed

Last 10 calls · auto-refreshing

Time	User / service	Model	Tokens	Status
10:42:11	[email protected]	gpt-4o-mini	412 in / 88 out	allowed
10:42:09	[email protected]	claude-3-5-sonnet	2,140 in / 540 out	redacted3 email addresses redacted
10:42:04	[email protected]	gpt-3.5-turbo	—	blockedModel not on approved list
10:41:58	support-bot	gpt-4o-mini	780 in / 220 out	allowed
10:41:55	[email protected]	claude-3-5-sonnet	1,022 in / 410 out	allowed
10:41:50	[email protected]	gpt-4o	320 in / 1,200 out	redactedPII pattern (credit card)
10:41:42	[email protected]	claude-3-haiku	215 in / 80 out	allowed
10:41:33	[email protected]	gpt-4-vision-preview	—	blockedModel not on approved list
10:41:28	summary-cron	gpt-4o-mini	1,400 in / 320 out	allowed
10:41:14	[email protected]	gpt-4o-mini	180 in / 76 out	redacted1 person name redacted

Redaction rules in force

Patterns matched server-side. GreyScape.ai itself never sees the raw input — redaction happens at the Cloudflare edge.

Email addresses11 caught
[email protected] → [EMAIL]
Person names9 caught
Maya Reyes → [PERSON]
Credit-card patterns4 caught
4242 4242 4242 4242 → [CARD]
Internal hostnames3 caught
prod-db-01.acme.local → [HOST]
AWS access keys1 caught
AKIA… → [AWSKEY]

3 policy blocks worth investigating

[email protected] tried to use gpt-3.5-turbo. Blocked — model is not on the approved list (banned for data-residency). Suggest gpt-4o-mini.
[email protected] tried to use gpt-4-vision-preview twice. Blocked — preview model not yet vendor-vetted. Suggest waiting for GA, or submit an approval request.

How you'd use this

Configure Cloudflare AI Gateway as the proxy endpoint for each approved provider (one-time).
Set redaction rules + approved-models policy in /settings/policies.
Watch the feed. Investigate blocks; tune redaction rules; export logs for compliance.

Sample output

12,847 AI calls inspected in 24h. 28 contained sensitive content and were redacted before leaving the network. 3 calls blocked entirely for using a banned model.

What this unlocks

Hard enforcement of the approved-models policy at the network edge. PII never leaves your boundary. Every AI call has a full audit record — proves to a customer auditor that prompts are inspected and policy is enforced, not just suggested.