On this page
- 01What you'll set up
- 02How a profile protects a request
- 03Create the profile
- 04Make it your organization baseline
- 05Choose what to protect
- 06Choose Redact or Block
- 07Add your own rules
- 08Add content safety
- 09Create it
- 10Test it with a real request
- 11Roll out with confidence
- 12What is checked
- 13If something goes wrong
01
What you'll set up
You will create a guardrail profile that keeps credentials out of every request your apps send to a model, make it your organization baseline, and test it with a real request.
- A profile with the Cloptima Secrets template
- Redact or Block for each template and rule
- Optional content safety, and what happens if its scan can't run
- The profile set as your organization baseline
- A test that shows what the model actually receives
You need an owner or admin role and an AI plan. Nothing is checked until you create a profile, so you decide exactly what is protected.
02
How a profile protects a request
A profile does two jobs. It protects sensitive data in every request before the model sees it, and it can check requests and responses for harmful content.
1Your app sends a request
2Protect sensitive data
Redact or block, before the model sees it
3Content safety
Optional, checks the request
4The model
5Content safety
Optional, checks the response
6Your app gets the answer
Redacted values come back
Data rules check the whole request: the system prompt, earlier turns, and the new message. A redacted value never reaches the model provider or a content-safety vendor, and your app still gets it back in the response, so your workflows keep working.
03
Create the profile
Open AI → Guardrails and choose New profile. Give the profile a clear name, such as Company baseline. The description is optional.
- Name
- Company baseline
- Description
- Optional
04
Make it your organization baseline
Switch on Org baseline and the profile applies to every policy. A policy can add one more profile for a team's own data, in the Guardrails field of its Basics section.
- Org baseline
- Applies to every policy and always enforces, even under a policy that observes. A block in either profile wins.On
| Baseline says | Team profile says | Result |
|---|---|---|
| Secrets: Redact | Project codenames: Block | Credentials are redacted everywhere. The codenames are blocked for that team. |
| US personal data: Redact | US personal data: Block | That team's personal data is blocked. A block in either profile wins. |
A team profile can add protection, and it can never weaken the baseline.
05
Choose what to protect
Under Protect sensitive data, switch on the Cloptima templates that fit your data. Each template is a ready set of rules that Cloptima maintains.
| Template | What it finds |
|---|---|
| Secrets | API keys and tokens from major cloud, AI, and developer platforms, private keys, bearer tokens, and values assigned to a password or key name |
| Source code secrets | Passwords inside connection URLs, and JSON Web Tokens |
| US, European, UK, Canadian, Indian, and Australian personal data | Email addresses, national ID numbers, phone numbers, and card numbers in each region's formats |
| Financial identifiers | SSNs, IBANs, and card numbers, while names, email addresses, and phone numbers keep flowing to your sales and CRM workflows |
| US financial data | Card numbers, bank routing and account numbers, IBANs, SWIFT codes, and crypto wallet addresses |
| US healthcare data | Medical record, health plan, and Medicare numbers, and other HIPAA identifiers |
| Infrastructure identifiers | IP and MAC addresses, AWS ARNs and account IDs, and internal host names |
Secrets is available on every plan with guardrails. The other templates, custom rules, and content safety come with AI Growth and Enterprise.
A template stays at the version you chose. When Cloptima publishes an improved version, the profile tells you, and you decide when to update.
06
Choose Redact or Block
Each template you switch on has its own action, and so does each custom rule. Redact is the default.
| Action | What happens | Use it for |
|---|---|---|
| Redact | The match becomes a placeholder before the model sees it. When the answer repeats the placeholder, your app gets the original value back. | Credentials and personal data your work still needs |
| Block | A match in the new message refuses the request with a clear reason | Data that must never reach a model |
Protect sensitive data
- Secrets
- Credentials with a known shape: cloud, AI, and source-control keys and tokens, private keys, bearer tokens, and values assigned to a password or key name.On
- Secrets action
- Redact
A match in an earlier turn of the conversation is always redacted, so a refused message never locks a conversation.
07
Add your own rules
Choose Add custom rule for data that only your business has, such as project codenames or employee IDs.
| Match | Finds |
|---|---|
| Term list | Fixed words and names, as whole words in any letter case |
| Regular expression (RE2) | Formats, such as account numbers and host names |
| Labeled value (RE2) | A value next to its label, such as Employee ID: E123456. Only the value is replaced. |
- Rule ID
- project_codenames
- Match
- Term list
- On match
- Block
- Terms (one per line)
- project falcon internal-only
Each term matches as whole words, in any letter case.
The guide on custom guardrail rules covers patterns, labeled values, and tuning in depth.
08
Add content safety
Switch on Content safety to check requests and responses for harmful content and block it. The check sees the text after your data rules redact it, so a vendor never sees a protected value.
| Implementation | You bring |
|---|---|
| Cloptima-managed (billed to AI credits) | Nothing. Cloptima runs the scan, and it draws from your AI credits. |
| Azure AI Content Safety | An Azure credential and the Content Safety endpoint |
| AWS Bedrock Guardrails | A Bedrock credential, the AWS region, and your guardrail ID and version |
| Google Model Armor | A Google service account and your Model Armor template |
| Webhook | Your HTTPS endpoint, with optional authentication |
Then choose If the scan can't run. Continue on local rules keeps traffic flowing on your data rules, and the request is listed as Not checked in Audit. Block the request refuses it until the scan runs again.
- Content safety
- Checks requests and responses for harmful content and blocks it. The check sees text after your data rules redact it. Content safety goes in the org baseline or in policy profiles, not both, so a request is checked once.On
- Implementation
- Cloptima-managed (billed to AI credits)
- If the scan can't run
- Continue on local rules
A request that continues is listed as Not checked in Audit.
09
Create it
Choose Create profile. The profile is live at once. Every save publishes a new version, and the version history and audit log record each change.
10
Test it with a real request
Send a prompt that contains a fake credential through your virtual key, and see what happens.
curl https://api.cloptima.ai/v1/ai/chat/completions \
-H "Authorization: Bearer $CLOPTIMA_VIRTUAL_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "gpt-5.4-mini", "messages": [{"role": "user", "content": "Debug this config: api_key = abc123def456ghi789"}]}'| Secrets action | What you see |
|---|---|
| Redact | A normal response. The model received a placeholder such as [SECRET_K7Q2M4XAB3CD] instead of the key, and any placeholder in the answer comes back to your app as the original value. |
| Block | HTTP 403 with the reason gateway_guardrail_blocked |
{
"error": "Your AI request was blocked because it matched a configured safety rule.",
"reason": "gateway_guardrail_blocked",
"violations": ["secret_assignment"]
}11
Roll out with confidence
A guardrail you trust is one you have watched work. Your policies' Monitor only mode lets you see a profile's effect before it changes any traffic.
- 1
Watch first
Attach the profile to a policy in Monitor only. The profile records what it would redact or block, and changes nothing. Requests it would have blocked appear in Audit, under Policy violations.
- 2
Redact
Switch the policy to Enforce. Credentials and personal data stop reaching models, and work continues.
- 3
Block what must never leave
Move the templates and rules that matter most to Block.
- 4
Make it the baseline
Switch on Org baseline when the profile fits every team.
12
What is checked
Data rules check every part of each request. Content safety checks the new message and the response, including streamed responses as they arrive.
Responses are never rewritten by your data rules. Each placeholder in a response is replaced with your original value before your app sees it.
The number of profiles and custom rules you can create depends on your plan.
13
If something goes wrong
Most surprises come from where a profile is attached, or from the policy's mode.
| What you see | Likely cause | Fix |
|---|---|---|
| Nothing is redacted or blocked | No baseline is set and the policy has no profile | Switch on Org baseline, or choose the profile in the policy's Guardrails field |
| The policy list shows No guardrails | No baseline exists and the policy has no profile | Create a baseline or attach a profile |
| Audit shows Would have blocked, and traffic is unchanged | The policy is in Monitor only | Switch the policy to Enforce, or make the profile the org baseline |
| Too much is blocked | A template or rule is set to Block | Set it to Redact while you tune |
| Requests show as Not checked | The content-safety scan could not run | Follow the fix Audit shows for the cause |