1. Overview
  2. Admin Guide
  3. Model Providers Overview

Model Providers Overview

Overview

Peridot supports multiple AI model providers and enables centralized control over how models are accessed, routed, and governed.

This allows organizations to use different models for different purposes while maintaining consistent policy enforcement and auditability.

Supported Providers

Peridot integrates with leading model providers:

  • OpenAI (GPT-4, GPT-4.1)

  • Anthropic (Claude 3.x)

  • Google (Gemini 1.5+)

  • Azure OpenAI

Organizations can also configure bring-your-own-key (BYOK) for full control over usage.

Why Multi-Model Support Matters

Different models have different strengths:

  • Reasoning quality

  • Cost efficiency

  • Latency

  • Data handling policies

Without centralized control:

  • Teams choose models independently

  • Policies are inconsistently applied

  • Risk increases

Peridot ensures all model usage is governed consistently.

Policy-Based Model Routing

Peridot routes requests dynamically based on policies.

Routing decisions can consider:

  • Data classification

  • User role

  • Environment

  • Application context

Example

  • Sensitive data → routed to internal or approved model

  • Low-risk requests → routed to cost-efficient model

  • High-complexity tasks → routed to advanced reasoning model

BYOK (Bring Your Own Key)

Peridot supports BYOK to ensure:

  • Full control over model usage

  • Data isolation

  • Compliance with internal policies

Keys are stored and managed within your environment.

Model Access Control

Access to models can be restricted by:

  • User roles

  • Environments

  • Applications

  • Data classification

This ensures that only approved systems can use specific models.

Multi-Step Model Workflows

Peridot supports workflows where different models are used at different stages.

Example:

  1. Retrieval model → fetch relevant data

  2. Reasoning model → process and generate output

  3. Execution model → trigger actions

This enables optimized performance and cost control.

Structured Outputs and Citations

Peridot supports:

  • Structured outputs (JSON, schemas)

  • Citations for traceability

This ensures outputs are:

  • Machine-readable

  • Auditable

  • Verifiable

Streaming Support

Peridot supports streaming responses for real-time applications.

This allows:

  • Faster user interactions

  • Improved user experience

  • Continuous output generation

Observability and Logging

All model interactions are logged, including:

  • Requests and responses

  • Model selection

  • Policy decisions

  • Execution outcomes

This enables:

  • Full auditability

  • Debugging and analysis

  • Compliance reporting

Risk Control

Peridot reduces model-related risk by:

  • Preventing unauthorized model usage

  • Enforcing data handling policies

  • Monitoring outputs and behavior

  • Detecting anomalies

In Production

In a production environment:

  • Model usage is centrally controlled

  • Routing is policy-driven

  • All activity is logged

  • Data exposure is minimized

Best Practices

  • Define approved models for each use case

  • Use policy-based routing instead of manual selection

  • Restrict external models for sensitive data

  • Monitor usage patterns regularly

Next Steps

Was this article helpful?