AI gateway
Send your AI calls through Valdra so it applies your rules, hides personal information and logs each call.
Read the video as text
AI gateway, in Valdra.
The clinic in this video is made up, and so is every detail.
Your own app may send text to an AI service. The gateway sits in between, checks each call against your rules, and logs it.
Start in Monitor mode: it logs what would be blocked and blocks nothing. Enforce blocks calls that break the policy.
Turn on redaction to mask personal information before the AI service sees it, then put it back in the answer.
The test console shows what a sample prompt would do, without calling any model. Use made up data only.
What this is for
Your own app may send text to an AI service. The gateway sits in between. It checks each call against your rules, hides personal information before the AI service sees it, and logs the call.
Before you start
- Plan: every plan can set up the policy and keys. The gateway takes calls only on Starter and up. On Free you will see The gateway runs on Starter and above.
- Role: only an Admin can change the policy, create or revoke keys and run the test. A Member or an Auditor can read the page.
- Have ready: the model ids your app uses. Your developer connects the app.
Steps
Open AI gateway
In the menu, open AI governance, then click AI gateway under Operate.

Choose the Enforcement mode
Under Governance policy, pick Monitor or Enforce. Monitor logs what would be blocked and blocks nothing. Enforce actually blocks calls that break the policy.
Start with Monitor to see what would happen.
Turn on the checks you want
Turn on Redact personal information to mask it before the model sees it. Turn on Limit to allowed models, Block threats or Data loss prevention as you need.
Under Allowed models, type the model ids, separated by commas. Under Banned terms, type words that must never be sent.
Click Save the policy. You will see Policy saved.
Test a prompt
In Policy test console, type the Model id and a sample Prompt. Use made up data only. Click Run the test.
You will see Allowed, Blocked or Would block (Monitor mode), and which checks fired. No model is called and no data is sent.
Connect your app
Under Gateway keys, type a key name and click Create a key. Copy the key at once. It will not be shown again.
Under Integration, click Copy and give the example to your developer.
What you will see
Once calls come in, tiles show the last 30 days: Calls (30 days), Blocked, Would block, Threats, Data loss hits and Personal information redacted.
Recent calls lists each call with its result. You can download the calls as CSV or JSON.
Common mistakes and fixes
You cannot change the policy. Only an Admin can. Ask an admin of your workspace.
You see View only instead of Run the test. Only an Admin can run the test. Ask an admin of your workspace.
Every call is refused. Gateway turned on is off, or your workspace is on Free.
Nothing is blocked in Monitor mode. That is expected. Switch to Enforce to block.
If it did not work
The policy was not saved. Try again. Click Save the policy again.
Policy not loaded Your saved policy could not be read, so saving is turned off. Reload the page.
The test failed. Try again. Click Run the test again.
Enter a model and a prompt to run the test. Fill in both fields.
The key was not created. Try again. Click Create a key again.
Some data did not load Part of the page could not be read. Reload the page.