User Guide
The AIM Starfort app inspects text for personal data with the AIM Intelligence Starfort Guard API. Starfort answers each inspection with one of three verdicts.
| Verdict | Meaning | Handling |
|---|---|---|
| PASS | No personal data, or only test values | Use the original as is |
| MASK | Personal data was found and hidden | Use processed_content (the masked text) instead of the original |
| BLOCK | The content must not be sent | Do not use it; stop |
A call that gets no verdict (HTTP 4xx, 5xx, timeout) is not a pass. The app always treats it as an error.
Before you start
A connect profile of type AIM Starfort must be registered. See the Install Guide.
Ways to use it
| Way | Purpose | Guide |
|---|---|---|
| Guardrail test screen | Send messages yourself and see the verdict and detections | Guardrail Test |
| AI agent guardrail | Automatically inspect what AI agents exchange with their models, and mask or block it | AI Agent Guardrail |
| Query commands | Inspect text or log records in a query | Query Commands |
Masking forms
Starfort masks each kind of personal data in its own way.
| Kind | Original | Masked |
|---|---|---|
| Mobile number | 010-1234-5678 | 010-****-5678 (middle digits hidden) |
hong.gildong@example.co.kr | [EMAIL_1] (replaced by its category) |
Values commonly used for testing, such as 010-0000-0000 and test@test.com, are treated as PASS.
Note
Starfort responses include the matched originals (
matched_text, original_text). They also appear on the guardrail test screen, in query command results and in the AI agent guardrail audit log, so control who can read them.
Contents
| Page | Contents |
|---|---|
| Guardrail Test | Building a request, the response panel, actions per error |
| AI Agent Guardrail | What is inspected, masking and restoring, failure handling, run details, the guardrail audit log |
| Query Commands | starfort-guard, starfort-guard-batch |
If this is your first time, check on the Guardrail Test screen that verdicts come back normally.