Profiles act as customizable rulesets that control how AI interactions are monitored and governed. They determine which signals to detect, how to respond to violations, and what actions to take on inputs and outputs.
Creating a profile is a matter of describing your AI deployment accurately. The wizard asks a series of questions about how your system is built and used, and it translates your answers into a starting guardrail configuration suited to that setup. The questions are not arbitrary preferences. Each one describes part of your deployment's risk surface, such as who interacts with it, what data it touches, an dhow much autonomy it has. Answer them as they truly reflect your system, because the resulting profile is only as well-fitted as the description it is built from.
The wizard has four steps, and a progress bar at the top of each step indicates how far along you are.
Step 1: Name and Type
1. Navigate to Inventory > Profiles and click Create Profile.
2. Enter a Profile Name. This field is required.
3. Optionally enter a Description.
4. Select a Deployment Type:
- Conversational: A chatbot or conversational AI that responds to user messages in a dialogue format.
- Agentic: An autonomous AI that takes actions, calls tools, and completes multi-step tasks.
5. If Agentic is selected, answer the follow-up question: Are the agent's actions reversible? This
describes whether the agent's actions can be undone. Irreversible actions carry higher stakes than
reversible ones, so the distinction informs how strictly the agent is governed.
- Reversible: Actions can be undone or rolled back, such as drafting emails or staging changes.
- Irreversible: Actions cannot be undone, such as sending payments, deleting data, or executing transactions.
Step 2: Users and System Configuration
Answer four questions about how the AI deployment operates.
1. Who are the users? Different audiences carry different risk. Traffic from external users generally
warrants closer scrutiny than internal use.
- Customers: External users interacting with your product or service.
- Employees: Internal team members using the AI as a productivity tool.
- Agents: Machine-to-machine only, with no humans interacting directly.
2. System prompt configuration: A locked prompt is a fixed, smaller attack surface, while a
configurable one can be altered at runtime and opens more room for manipulation.
- Locked: The system prompt is fixed and cannot be changed at runtime.
- Configurable: The system prompt can be modified by users or at runtime.
3. Does the AI ingest external content? Content pulled in from outside sources can carry injected
instructions or sensitive data, so it expands what the profile needs to watch for. External content
includes web pages, uploaded documents, emails, or any user-provided data that gets embedded
into the context. Select Yes or No.
4. Does the AI use persistent memory? When information carries across sessions, data can persist and
resurface later, which raises data-handling considerations a stateless deployment does not have.
Persistent memory retains information across sessions, such as user preferences, conversation
history, or learned context. Select Yes or No.
Step 3: Sensitive Data Types
Select any sensitive data types that may be present in AI interactions. This step is optional.
- PII: Personally identifiable information, including names, addresses, and SSNs.
- PCI: Payment card data, including credit card numbers, CVVs, and cardholder names.
- PHI: Protected health information, including medical records, diagnoses, and treatment data.
- Secrets: API keys, tokens, passwords, and other credentials.
- Customer Data: Business customer data that requires special handling.
Step 4: Output and Review
1. Select how AI output reaches end users.
- Direct: AI responses are shown directly to end users without human review.
- Reviewed: A human reviews AI output before it reaches end users.
2. Review the Review Your Choices summary, which displays all selections before the profile is created:
Name, Deployment Type, Reversibility, Users, System Prompt, External Content, Persistent Memory,
Sensitive Data, and Output.
3. Click Create Profile to complete the flow.