Concepts

WitnessAI helps you observe, control, and protect how AI is used across your organization. This page explains the core building blocks and how they relate, so the rest of the documentation makes sense. Each section links to the page with full detail.

Policies

A policy is the rule that decides how WitnessAI governs an AI interaction. Each policy specifies who and what it applies to and which guardrails inspect the traffic. WitnessAI evaluates policies top-down and applies the first one that matches an interaction. See Create a policy and Policy precedence.

Sources and destinations

A policy applies only when an interaction matches both a source and a destination.

Source: who the interaction comes from: any user, specific users, or groups.
Destination: where the interaction goes: any provider, specific AI providers, models, applications, your private applications, or a reusable list.

A policy triggers only when both sides match, which lets you scope rules precisely. See Create a policy.

Guardrails

A guardrail is an inspection plus an action that runs on traffic a policy matches. Guardrails detect things like sensitive data, risky topics, harmful responses, and unusual behavior, and they can allow, warn, block, route, or anonymize the activity. You attach one or more guardrails to a policy. See the guardrail pages under Guardrails, starting with Data Protection and Risk Analysis.

Discovery

Discovery gives you organization-wide visibility into the AI in use: which applications, agents, and Model Context Protocol (MCP) servers people reach, how they are accessed, and the risk each carries. Use it to understand your AI footprint before and after you put policies in place. See Discovery, Agents, and MCP servers.

Anonymization

Anonymization keeps sensitive data away from AI models without breaking the user experience. When enabled in the Data Protection guardrail, WitnessAI replaces detected sensitive values with placeholders before the prompt reaches the model, then restores the original values in the response. See Data Protection.

Secure AI Portal

The Secure AI Portal is a protected, chat-style interface for talking to AI models, similar to a consumer chat tool. Every conversation runs through your organization's policies and guardrails, so it is a fast way to give users governed AI access with no client setup. See Secure AI Portal.

Witness Anywhere

Witness Anywhere extends WitnessAI coverage to employees working outside the corporate network. It enforces your AI policies at the endpoint, so remote and off-network devices stay governed even when their traffic does not route through enterprise infrastructure. See Witness Anywhere overview.

Alerts and analytics

Once traffic flows through WitnessAI, the console turns it into insight:

Conversations show the prompts and responses captured across your organization. See Conversations.
Alerts surface the events your policies and guardrails act on, with full match detail. See Alerts.
Analytics summarize usage and risk trends over time. See Analytics.

Where things are in the console

The console's left navigation has Home, Discovery, Analytics, Conversations, Users, Alerts, and a Policy section with Policies and Lists. Administrators also see an Administration section. The Apps & Resources menu at the bottom of the sidebar links to the Secure AI Portal.

Discovery includes Discovered Apps, App Catalog, MCP Servers, and Agents.
Analytics groups its dashboards into Overview, Guardrails, and Platform. Under Platform, the dashboards include Witness Anywhere, SOC Dashboard, and ROI.
Administration has Configuration (API Keys, Models, Proxy, and SIEM), Access Control (Users, Groups, and Roles), and System (Audit Logs). The Proxy page's Witness Anywhere section covers managed devices in unmanaged environments.

To protect your own applications, see the API Reference for the Guardrails API.

How it fits together

In practice, the pieces form a loop: Discovery shows you what AI is in use, policies (with their sources, destinations, and guardrails) define how to govern it, traffic reaches WitnessAI through the Portal, Witness Anywhere, a network integration, or the API, and Conversations, Alerts, and Analytics show you the result so you can refine your policies.

Concepts Policies Sources and destinations Guardrails Discovery Anonymization Secure AI Portal Witness Anywhere Alerts and analytics Where things are in the console How it fits together