How Peridot Works

Peridot is designed as a control layer that operates across your AI systems, data sources, and integrations.

It separates control from execution, allowing you to govern AI usage without disrupting how teams build and use AI.

Architecture overview

Peridot uses a dual-layer architecture:

Control Plane (Peridot Cloud)

  • Policy definition and enforcement

  • Metadata and configuration

  • Audit logs and traceability

  • Workflow orchestration

Data Plane (Customer Environment)

  • Data processing

  • Model inference (via BYOK or customer-controlled access)

  • Integration execution

  • Application runtime

This separation ensures that sensitive data remains within your environment while control logic remains centralized.

The control flow

Peridot operates across a continuous lifecycle:

1. Discover

Peridot identifies all AI tools and systems in use across your environment.

This includes both sanctioned and shadow AI usage.

2. Monitor

Peridot tracks how data flows into and out of AI systems.

This includes:

  • What data is being used

  • Which models are involved

  • Where outputs are sent

3. Control

Policies are applied to govern behavior.

This includes:

  • Model access controls

  • Data restrictions

  • Integration permissions

  • Approval requirements

4. Enforce

Peridot enforces policies in real time.

This may include:

  • Blocking actions

  • Re-routing model usage

  • Triggering approvals

  • Logging violations

5. Respond

When issues occur, Peridot creates incidents.

These incidents can trigger:

  • Alerts

  • Workflows

  • AI incident playbooks

6. Audit

All activity is recorded and traceable.

This includes:

  • Inputs and outputs

  • Policy decisions

  • Model usage

  • Actions taken

Model control

Peridot supports:

  • Multiple model providers

  • Bring your own keys (BYOK)

  • Policy-based routing

  • Structured outputs and citations

Model usage is governed by policy, not by individual application configuration.

Why this matters

Without a control layer:

  • AI usage is fragmented

  • Data exposure is unmanaged

  • Policies are unenforced

  • Auditing is incomplete

Peridot centralizes control without slowing down AI adoption.

What to do next

  • Start with AI Inventory Overview to discover AI usage

  • Define your first Policies to begin enforcement


Was this article helpful?