1. Overview
  2. Builder Guide
  3. Policy-Based Routing

Policy-Based Routing

Overview

Policy-based routing determines how AI requests are directed across models and providers.

Instead of relying on application-level logic, Peridot uses centralized policies to dynamically select the appropriate model for each request.

Why Routing Matters

Without routing:

  • Model selection is inconsistent

  • Sensitive data may be sent to unapproved providers

  • Cost and performance are not optimized

  • Governance is fragmented

With routing:

  • Model selection is centralized

  • Policies enforce compliance automatically

  • Requests are routed safely and consistently

How Routing Works

Every AI request follows a controlled routing path:

  1. Request is received

  2. Policies are evaluated

  3. Constraints are applied

  4. An approved model/provider is selected

  5. The request is routed and executed

Routing Inputs

Routing decisions are based on:

  • Data sensitivity

  • User role

  • Application context

  • Environment

  • Cost and performance constraints

Common Routing Patterns

Sensitive Data Routing

Sensitive data is routed to internal or approved models only.

Cost Optimization

Low-risk requests are routed to lower-cost models.

Environment-Based Routing

  • Development → sandbox models

  • Production → approved models

Provider Restrictions

Requests are limited to approved providers (e.g. Azure OpenAI only).

Multi-Step Routing

Peridot supports routing across multi-step workflows:

  • Retrieval → internal system

  • Reasoning → approved model

  • Execution → API or integration

Each step is governed independently.

How It Works in Production

In production:

  • Routing decisions are applied in real time

  • Changes do not require application updates

  • Policies ensure consistent behavior across all systems

  • All routing decisions are logged and auditable

Benefits

  • Centralized control over model usage

  • Reduced risk of data exposure

  • Improved cost and performance management

  • No dependency on application-level logic

Best Practices

  • Define routing policies early

  • Separate dev and production routing

  • Combine routing with enforcement actions

Next Steps

Was this article helpful?