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

# Arya AI settings

> Configure AI provider, models, generation parameters, BYOK.

Configure how Arya uses AI: which provider, which model, generation parameters, cost limits.

## AI provider

<Tabs>
  <Tab title="Arya managed (default)">
    Arya manages everything: infrastructure, providers (mix of Anthropic Claude, OpenAI GPT, Google Gemini), automatic routing to best model. Costs included in credits.
  </Tab>

  <Tab title="BYOK (Bring Your Own Key)">
    Use your API key (OpenAI, Anthropic, Azure OpenAI). Costs directly with provider, Arya doesn't charge AI credits.
  </Tab>

  <Tab title="Private deployment (Enterprise)">
    Dedicated models on private infrastructure (AWS/Azure/on-prem). For data residency or performance needs.
  </Tab>
</Tabs>

## Configure BYOK

<Steps>
  <Step title="Settings → AI → Provider → BYOK">Pick OpenAI or Anthropic.</Step>
  <Step title="Paste API key">Generated from provider portal.</Step>
  <Step title="Test">Arya sends test request. If works, save.</Step>

  <Step title="Pick model">
    OpenAI: gpt-4.1, gpt-4o, o3, o4-mini, etc.
    Anthropic: claude-opus-4, claude-sonnet-4-6, claude-haiku-4-5, etc.
  </Step>
</Steps>

## Parameters

<AccordionGroup>
  <Accordion title="Temperature">
    0 = deterministic, 1 = creative. Default 0.3 (balanced).

    * Low for structured data, classifications
    * High for brainstorming, creativity
  </Accordion>

  <Accordion title="Max tokens">
    Max output length. Default 2000.
  </Accordion>

  <Accordion title="Reasoning level (compatible models only)">
    How much "thinking" before responding. High = more accurate but slower and more expensive.
  </Accordion>

  <Accordion title="Verbose mode">
    Shows step-by-step reasoning. Useful for debugging, superfluous for normal use.
  </Accordion>
</AccordionGroup>

## Cost limits

To avoid unexpected consumption:

<Steps>
  <Step title="Monthly budget">Credit threshold. Beyond → Arya blocks non-essential requests.</Step>
  <Step title="Per-user budget">Limit individual consumption.</Step>
  <Step title="Threshold alerts">70% quota → email; 90% → notification; 100% → block.</Step>
</Steps>

## Activatable AI features

Pick what Arya can do with AI:

<CardGroup cols={2}>
  <Card title="AryaChat">Conversational chat.</Card>
  <Card title="Document generation">Emails, contracts, quotes.</Card>
  <Card title="Call intelligence">Transcription + call summary.</Card>
  <Card title="Report chat">Natural language data analysis.</Card>
  <Card title="Data enrichment">Company and contact enrichment.</Card>
  <Card title="Receipt OCR">Expense reports from photos.</Card>
  <Card title="Auto-tagging">Automatic record classification.</Card>
  <Card title="Smart suggestions">Proactive suggestions (next-best-action).</Card>
</CardGroup>

Disable what you don't need to save credits.

## AI privacy

<Note>
  * **Arya managed**: GDPR compliant, no use for global model training
  * **BYOK OpenAI/Anthropic**: subject to provider policies. Verify zero-retention if you have sensitive data.
  * **Private deployment**: data never leaves your infrastructure.
</Note>

## Monitor usage

Dashboard **Settings → AI → Analytics**:

* Consumption per feature
* Consumption per user
* Time trend
* Most expensive queries (for optimisation)

## Frequently asked questions

<AccordionGroup>
  <Accordion title="Which provider is best?">Depends on use case. Claude excels at long reasoning and analysis. GPT at creativity and code. Arya managed auto-routes to best for each task.</Accordion>
  <Accordion title="Can I use open-source models (Llama, Mistral)?">On Enterprise plans with private deployment, yes. On standard plans, no.</Accordion>
</AccordionGroup>