The Profile Detail page is where you see and control how a single profile evaluated and acts on you AI traffic. From here you set the model and mode the profile runs in, choose which signals it detects, and configure what happens when each one fires. You reach the page by selecting a profile from the Profiles list.

The header keeps you oriented while you work: it shows the profile's name, status, current mode, and how many signals are enabled, along with when it was last changed and by whom. From the top of the page you can open the profile's monitoring view, or use the menu to duplicate or delete the profile.

Model and Mode

The most consequential choice on the page is the model and mode, because it determines whether the profile acts on traffic or only observes it.

A profile runs in one of two modes:

• Listen analyzes traffic in parallel and returns results alongside suggested actions, without changing the traffic. Results are available through the API or in Monitoring.
• Enforce places the profile in-line between your application and the model, acting on input before it reaches the LLM and on the response before it returns.

Enforce requires a model, so you select on before the profile can enforce. A profile that has no model selected can only run in Listen mode. Once a model is set, the section shows the model, its provider, and its context window.

Configuring Detection

A profile detects traffic through signals. Each signal can be turned on or off, and a signal is only in effect when it is enabled. The header's enabled-signal count reflects how many are currently active.

For signals that detect specific content, you control what happens when a detection occurs. Detection is configured separately for the prompt and the response, so a signal can be handled differently depending on which side it appears on. When a detection occurs, the profile can:

• Pass it through unchanged, recording it for observability without altering the content.
• Modify the content, redacting or rewriting the matched portion before passing it on.
• Block the content, rejecting the prompt or response outright.

Detections can also be forwarded to your configured SIEM destinations. Some detections apply only to the prompt side.

What a Profile Can Detect

A profile offers sixteen signals, spanning the kinds of risk and content that matter across AI traffic. They fall into a few broad areas:

• Model and processing: Model sets the backing model and mode, Semantic Chunking governs how longer content is segmented for analysis, and Tools surfaces tool calls in event details.
• Data: Named Entities, PII / PHI / PCI, and Secrets detect sensitive, personal, and credential data in the content.
• Safety and conduct: Toxicity and Illegality detect harmful, hostile, or unlawful material, and Sentiment classifies the tone of the content.
• Code: Code Requested and Code Present detect requests to generate cod and cod already present in the content, and Code Vulnerability covers insecure code.
• Adversarial and agentic: Adversarial detects attempts to manipulate or exploit the model, and Agentic Instructions detects agent instruction and planning content.
• Intent and language: Intent classifies what the use is trying to do, and Languages classifies the language of the content.

Each signal has its own settings, and selecting a signal opens its configuration, where you enable it and set how its detections are handled on the prompt and response.