> ## Documentation Index
> Fetch the complete documentation index at: https://docs.praxis-ai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Switching to Praxis AI Middleware

> Unlock the full power of AI orchestration with enterprise-grade flexibility

## Why Switch to Praxis-AI Middleware?

Praxis-AI middleware transforms how organizations leverage AI by providing a sophisticated orchestration layer that connects your existing systems with multiple AI models while maintaining complete control, security, and flexibility.

<CardGroup cols={3}>
  <Card title="Bring Your Own Token (B.Y.O.T.)" icon="shuffle">
    Seamlessly switch between AI models such as OpenAI, Anthropic Claude, and Google Gemini without vendor lock-in. Use your existing API keys and credit pools across all major AI platforms.
  </Card>

  <Card title="Enterprise Security" icon="shield-halved">
    Bank-level encryption, SOC 2 compliance, FERPA & COPPA adherence, complete data sovereignty with zero proprietary data exposure
  </Card>

  <Card title="One Knowledge Base, Every Model" icon="chart-line">
    Your IP Vault documents, assistants, tools, and conversation history work the same no matter which AI model answers — switch providers without rebuilding anything
  </Card>

  <Card title="Cost Optimization" icon="dollar-sign">
    Prompt caching discounts (cached input tokens are billed at a reduced rate), per-task model choices, and transparent credit accounting help you control AI costs
  </Card>
</CardGroup>

***

## Bring Your Own Token (B.Y.O.T.)

<Info>
  Switching to Praxis-AI middleware is seamless. Use your existing API keys and credit pools across all major AI platforms.
</Info>

### Supported AI Platforms

Praxis-AI middleware provides native integration with industry-leading AI platforms:

<AccordionGroup>
  <Accordion title="OpenAI Models" icon="brain">
    Connect your OpenAI API key to access:

    * GPT-5, GPT-5-mini for generation
    * DALL-E 3 for image generation
    * Whisper for speech-to-text

    **Configuration:** Simply add your API key in Edit → Configuration and Integrations
  </Accordion>

  <Accordion title="AWS Bedrock Platform" icon="aws">
    Leverage your AWS Bedrock access for:

    * **Anthropic**: Claude Sonnet 4, Claude Sonnet 4.5, Claude Haiku 4.5
    * **Amazon Nova**: Premier, Pro, Lite, Micro
    * **Meta Llama**: Llama 4x
    * **Mistral AI**: Mistral Large
    * **Amazon Titan**: Text, Embeddings, Multimodal

    **Configuration:** Authenticate via AWS IAM credentials or access keys
  </Accordion>

  <Accordion title="Google AI Platform" icon="google">
    Integrate with Google's AI ecosystem:

    * Gemini 3.1 Pro and the Gemini Flash family
    * Gemini Live (real-time voice)
    * Vertex AI models

    **Configuration:** Add your Google AI API key in Edit → Configuration and Integrations
  </Accordion>

  <Accordion title="Mistral AI" icon="wind">
    Access Mistral's high-performance models:

    * Mistral Large, Mistral Medium
    * Codestral (code-focused)
    * Voxtral (voice TTS/STT)

    **Configuration:** Add your Mistral API key in Edit → Configuration and Integrations
  </Accordion>

  <Accordion title="xAI (Grok)" icon="bolt">
    Access xAI's Grok models:

    * Grok-4.20, Grok-4 Fast for conversation and vision
    * Grok Code Fast for code-focused tasks
    * Grok Imagine for image generation
    * Text-to-Speech (5 voices: Eve, Ara, Rex, Sal, Leo)
    * Embeddings (grok-embedding-small)
    * Real-time voice conversations (Grok-3 Fast)

    **Configuration:** Add your xAI API key in Edit → Configuration and Integrations
  </Accordion>

  <Accordion title="Stability AI" icon="image">
    Access Stability AI's generative media models:

    * **Image Generation:** Stable Image Ultra, Stable Image Core, SD 3.5 Large
    * **Audio Generation:** Stable Audio 2 (text-to-audio, up to 190 seconds)

    Stability AI is a dedicated media-generation provider — conversation models from other providers invoke the `generate_image` and `generate_audio` tools that route to Stability automatically. (Video generation routes to Amazon Nova Reel or OpenAI Sora.)

    **Configuration:** Add your Stability AI API key in Edit → Configuration and Integrations
  </Accordion>

  <Accordion title="Custom Models & Endpoints" icon="code">
    Connect to any AI model via REST API:

    * Self-hosted models (Ollama, vLLM, TGI)
    * LiteLLM proxy endpoints
    * Azure OpenAI Service
    * Custom fine-tuned models

    **Configuration:** Provide endpoint URL + API key/bearer token
  </Accordion>
</AccordionGroup>

***

## How It Works

<Steps>
  <Step title="Configure Your Models" icon="gear">
    Navigate to **Configuration → Personalization and AI Models** and add your API credentials for each platform you want to use.

    <Tip>
      You can configure multiple models simultaneously and switch between them based on use case, cost, or performance requirements.
    </Tip>
  </Step>

  <Step title="Set Model Preferences" icon="sliders">
    Define which models to use for different Digital Twins or conversation types:

    * **High-reasoning tasks**: Claude Sonnet 4.6, GPT-5.2, Gemini 3.1 Pro
    * **Fast responses**: Claude Haiku 4.5, Gemini Flash, Amazon Nova Lite
    * **Cost-sensitive operations**: Amazon Nova Micro, GPT-5-mini
    * **Specialized domains**: Custom fine-tuned models
  </Step>

  <Step title="Assign Models per Task" icon="route">
    Each Digital Twin can use a different model for each kind of work — conversation, summaries, image generation, image analysis, audio transcription, video analysis, embeddings, and real-time voice. Assistants can also pin their own conversation model, overriding the Twin's default.
  </Step>

  <Step title="Monitor & Optimize" icon="chart-line">
    Track usage, token consumption, and credit spend across all models through the admin Analytics and Histories dashboards, then adjust your model assignments to optimize for your specific needs.
  </Step>
</Steps>

***

## Key Advantages Over Direct API Integration

<CardGroup cols={2}>
  <Card title="Unified Interface" icon="layer-group">
    A single API and a single admin console for all AI models — no per-provider integration work
  </Card>

  <Card title="Easy Model Switching" icon="arrows-rotate">
    If a model is deprecated, unavailable, or too expensive, switch to an alternative in the admin console — no code changes
  </Card>

  <Card title="Context Management" icon="memory">
    Persistent conversation memory and RAG vector search across all models with institutional knowledge prioritization
  </Card>

  <Card title="Token Optimization" icon="gauge-high">
    Prompt caching discounts on supported models and per-task model choices reduce the cost of repeated context
  </Card>

  <Card title="Compliance Built-In" icon="file-shield">
    Content moderation, conversation audit history, and granular access controls support your regulatory obligations
  </Card>

  <Card title="No Vendor Lock-In" icon="unlock">
    Switch models instantly without code changes or data migration
  </Card>
</CardGroup>

## Migration Path

<Note>
  Switching from direct API integration to Praxis-AI middleware requires minimal changes to your existing infrastructure.
</Note>

### Three-Step Migration

1. **Add Praxis-AI SDK**: Install the middleware SDK alongside your existing AI integrations
2. **Configure Models**: Import your existing API keys into Praxis-AI's model configuration
3. **Update Endpoints**: Point your application to Praxis-AI's unified API endpoint

### Zero Downtime Migration

Run Praxis-AI middleware in parallel with your existing integration during the transition period. Gradually shift traffic to validate performance before full cutover.

***

## Enterprise Features

<AccordionGroup>
  <Accordion title="Model Context Protocol (MCP) Integration" icon="plug">
    Praxis-AI implements the Model Context Protocol standard both ways: expose your Digital Twin as an MCP server to Claude Desktop, Cursor, or any MCP-compatible client, and connect your Twin to remote MCP servers to extend its tool set. See [MCP Server](/mdx/integrations/mcp/introduction).
  </Accordion>

  <Accordion title="IP Vault & Knowledge Management" icon="vault">
    Secure storage for institutional knowledge with hierarchical access:

    1. LMS Content such as Canvas
    2. IP Vault proprietary content
    3. Trusted external sources

    All content is encrypted and access-controlled with role-based permissions.
  </Accordion>

  <Accordion title="Digital Twin Architecture" icon="users">
    Create sophisticated AI experts that preserve human reasoning patterns and institutional knowledge. Each Digital Twin can use different models optimized for their specific domain.
  </Accordion>

  <Accordion title="Advanced Analytics" icon="chart-area">
    Comprehensive admin dashboards tracking:

    * Token usage and credit spend, broken down per model
    * Conversation activity, sessions, and engagement over time
    * Per-user usage statistics with CSV export
    * User feedback with admin responses
  </Accordion>
</AccordionGroup>

***

## Pricing & Cost Optimization

<Tip>
  Praxis-AI middleware adds minimal overhead while providing significant cost savings through intelligent routing and token optimization.
</Tip>

### Cost Structure

* **Middleware Fee**: Credit-based system — pay as you go, or buy bundles and save
* **AI Model Costs**: Customers who bring their own API keys (BYOK) receive a 40% discount on standard pricing
* **No Hidden Fees**: Transparent pricing with volume discounts available

### Where the Savings Come From

* **Bring Your Own Keys** — a 40% discount on standard pricing when you supply your own provider credentials (see [Plans and Credits](/mdx/introduction/plans-and-credits))
* **Prompt caching** — cached input tokens are billed at a reduced rate on models that support caching, with the realized discount shown on each conversation's report card
* **Right-sized models** — assign inexpensive models to high-volume background work (summaries, embeddings) and reserve premium models for conversation
* **No failed-call charges** — AI calls that fail are not billed

***

## Getting Started

<Steps>
  <Step title="Sign Up" icon="user-plus">
    Create your Praxis-AI account at [https://pria.praxislxp.com](https://pria.praxislxp.com) or through AWS Marketplace
  </Step>

  <Step title="Create Digital Twin" icon="robot">
    Set up your first Digital Twin and run the personalization assistant in Configuration → Personalization and AI Models
  </Step>

  <Step title="Configure First Model" icon="wrench">
    Configure your Digital Twin to use your API Token
  </Step>

  <Step title="Test & Deploy" icon="rocket">
    Test conversations and deploy to your organization via LMS integration, Web SDK, or REST APIs
  </Step>
</Steps>

<Card title="Need Help?" icon="headset" href="/mdx/contact-support">
  Contact our team for personalized onboarding assistance and technical support
</Card>

***

## Additional Resources

<CardGroup cols={3}>
  <Card title="Choose a model" icon="sliders" href="/mdx/admin-guide/ai-models">
    Personalize your Digital Twin models
  </Card>

  <Card title="Configure your API Keys" icon="sliders" href="/mdx/admin-guide/configuration">
    API Keys Configuration for your Digital Twin
  </Card>

  <Card title="Integrate your Digital Twin" icon="book" href="/mdx/integrations/overview">
    Integrate your Digital Twin
  </Card>

  <Card title="Connect via our APIs" icon="book" href="/mdx/api-reference/introduction">
    Complete API reference and integration guides
  </Card>

  <Card title="Adding Credits" icon="coins" href="/mdx/user-guide/credits">
    Adding credits in 3 steps
  </Card>
</CardGroup>