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

# AI trust & privacy

> How Ravenna AI agents handle your data: encryption, model provider policies, training opt-outs, audit logs, and admin controls over AI features.

Ravenna AI <Tooltip headline="Agents" tip="AI assistants that automate support and answer questions" cta="Learn about agents" href="/documentation/automate/agents/overview">agents</Tooltip> process your data to answer questions, analyze tickets, and automate workflows. All AI processing uses enterprise-grade security with no training on your data.

<AccordionGroup>
  <Accordion title="Data protection" defaultOpen>
    All data encrypted in transit and at rest. AI model providers (Anthropic, OpenAI) do not train on your data or retain prompts and responses. SOC 2 Type II certified.
  </Accordion>

  <Accordion title="Your control">
    Workspace admins control which channels have AI agents, restrict <Tooltip headline="Knowledge base" tip="Searchable documentation and resources" cta="Learn about knowledge" href="/documentation/automate/knowledge/overview">knowledge base</Tooltip> access, and review AI usage logs. Users can request AI not respond to their messages.
  </Accordion>

  <Accordion title="Transparency">
    AI responses cite knowledge base sources. Conversation logs help you monitor quality and accuracy. Clear limitations documented to set appropriate expectations.
  </Accordion>
</AccordionGroup>

***

## Understanding AI processing

### How Ravenna uses AI

AI features automate support tasks and provide intelligent assistance:

* **Answer questions** using your knowledge base articles and documentation
* **Create tickets** with pre-filled <Tooltip headline="Forms" tip="Structured intake forms with custom fields" cta="Learn about forms" href="/documentation/tickets/forms/overview">forms</Tooltip> based on conversation context
* **Trigger workflows** to automate approval processes and provisioning
* **Classify requests** to route tickets to appropriate teams
* **Generate insights** from ticket patterns and trends
* **Execute tool actions** through integrated services like Fleet or HubSpot

### What data is processed

AI features process data necessary to provide intelligent responses:

**Ticket data:**

* Ticket titles, descriptions, and comments
* Ticket metadata (status, priority, assignee)
* Historical ticket patterns and resolution data

**Knowledge base content:**

* Articles and documentation you've connected to agents
* Document metadata and structure

**Conversation data:**

* User messages in connected Slack channels
* Message context and thread history
* User mentions and reactions

**Integration data:**

* Tool execution results from connected integrations
* External system data accessed through tool actions

<Info>
  Only data explicitly connected to agents (through knowledge bases, channels, or tool permissions) is processed by AI. Agents cannot access data outside their configured scope.
</Info>

### Data flow

AI processing follows a secure data flow:

1. **User sends message** in Slack or creates ticket
2. **Agent retrieves context** from configured knowledge bases and conversation history
3. **Data sent to AI model** (Anthropic Claude, OpenAI, or Google Vertex AI) with relevant context
4. **Model generates response** using provided context without accessing external data
5. **Response delivered** to user with citations to knowledge base sources
6. **Conversation logged** in Ravenna for monitoring and debugging

<Note>
  AI models process data in real-time and do not retain your data after generating responses. All API calls use encrypted connections.
</Note>

### Sensitive information handling <Badge icon="bot" color="gray" shape="rounded">V3+</Badge>

When an agent's response contains sensitive information like passwords, API keys, tokens, or credentials, the agent automatically sends that content as a private direct message instead of posting it in the shared channel.

<AccordionGroup>
  <Accordion title="What other participants see" defaultOpen>
    The shared channel displays a placeholder message indicating that a private message was sent. Other participants in the thread do not see the sensitive content.
  </Accordion>

  <Accordion title="What the requester sees">
    The requester receives an ephemeral DM in Slack containing the actual sensitive information. This keeps credentials and secrets visible only to the person who requested them.

    <Note>
      Ephemeral secret DMs are Slack-only today. The Microsoft Teams integration does not deliver secrets via DM in the current beta. See the [Microsoft Teams integration overview](/integrations/microsoft-teams/overview) for beta limitations.
    </Note>
  </Accordion>

  <Accordion title="When private messages are triggered">
    Agents send private messages in two situations:

    * **Automatic detection:** The agent recognizes that its response contains passwords, API keys, tokens, connection strings, or other secrets from your knowledge base
    * **User request:** The user explicitly asks for a private or DM response

    You can also configure agents to send private messages for specific scenarios using <Tooltip headline="Rules" tip="Natural language instructions defining agent behavior" cta="Learn about rules" href="/documentation/automate/agents/configure#rules">rules</Tooltip>.
  </Accordion>
</AccordionGroup>

***

## Data protection

### Security measures

<CardGroup cols={2}>
  <Card title="Encryption" icon="lock">
    All data encrypted in transit using TLS 1.3 and at rest using AES-256 encryption
  </Card>

  <Card title="Access controls" icon="key">
    Role-based permissions control AI feature access at workspace and channel levels
  </Card>

  <Card title="Audit logs" icon="clipboard-list">
    Complete audit trail of AI interactions, tool executions, and configuration changes
  </Card>

  <Card title="Compliance" icon="badge-check">
    SOC 2 Type II certified with GDPR and CCPA compliance measures
  </Card>
</CardGroup>

### Data retention

**Conversation logs:**

* Stored in Ravenna for debugging and quality monitoring
* Retained according to your organization's data retention policy
* Can be exported or deleted upon request

**AI model provider data:**

* Prompts and responses not retained by model providers
* No training on your data by Anthropic, OpenAI, or Google
* API calls processed in real-time without storage

**Knowledge base content:**

* Stored in Ravenna and synchronized with connected sources
* Not sent to AI models unless explicitly referenced by agents
* Removed when knowledge base connection is deleted

### Data processing agreements

Ravenna maintains Data Processing Agreements (DPAs) with AI model providers:

* **Anthropic** (Claude AI models)
* **OpenAI** (GPT models)
* **Google** (Vertex AI)

These agreements ensure data is processed securely and in compliance with privacy regulations including GDPR, CCPA, and other regional requirements. DPAs specify that providers do not train models on your data or retain data after processing.

<Callout icon="link" color="#6B7280">
  Review the full [Ravenna Privacy Policy](https://ravenna.ai/privacy) for complete details on data handling and compliance
</Callout>

***

## Control & permissions

Configure AI features at multiple levels to control data access and agent behavior.

<AccordionGroup>
  <Accordion title="Workspace-level controls" defaultOpen>
    Workspace admins control AI features for the entire workspace:

    * Enable or disable AI agents globally
    * Configure which <Tooltip headline="Channels" tip="Organized workspaces for managing tickets by team or topic" cta="Learn about channels" href="/documentation/tickets/channels">channels</Tooltip> have AI agents deployed
    * Restrict knowledge base access to specific agents
    * Review AI usage logs and conversation history
    * Configure auto-respond settings for email and integration tickets
    * Manage tool permissions for agent integrations
  </Accordion>

  <Accordion title="Channel-level controls">
    Each channel can have independent AI configuration:

    * Deploy specific agent or no agent
    * Connect distinct knowledge base folders
    * Configure custom <Tooltip headline="Escalation instructions" tip="Guidelines for handling complex requests" cta="Learn about escalation" href="/documentation/automate/agents/configure#escalation-instructions">escalation instructions</Tooltip>
    * Set channel-specific response behavior
    * Control which <Tooltip headline="Rules" tip="Natural language instructions defining agent behavior" cta="Learn about rules" href="/documentation/automate/agents/configure#rules">rules</Tooltip> apply to the agent
  </Accordion>

  <Accordion title="User-level privacy">
    Individual users have privacy controls:

    * See which AI features are active in their channels
    * Request AI not respond to their messages
    * Provide feedback on AI responses (thumbs up/down)
    * Request data deletion for their conversations
    * View AI response sources and citations
  </Accordion>
</AccordionGroup>

<Info>
  Changes to agent configuration, knowledge base connections, and permissions are logged in audit trails accessible to workspace admins.
</Info>

***

## Responsible AI usage

### Limitations

AI agents have limitations you should understand before deployment:

<Warning>
  **Known limitations:**

  * May generate incorrect information or hallucinate facts not present in knowledge bases
  * Cannot access real-time external data without configured tool integrations
  * Limited by quality and completeness of connected knowledge base content
  * Cannot perform actions requiring human judgment, compliance review, or legal interpretation
  * May misinterpret ambiguous requests or questions lacking sufficient context
  * Cannot guarantee 100% accuracy even when citing knowledge base sources
</Warning>

<Info>
  Configure <Tooltip headline="Escalation instructions" tip="Guidelines for handling complex requests" cta="Learn about escalation" href="/documentation/automate/agents/configure#escalation-instructions">escalation instructions</Tooltip> to route complex, sensitive, or ambiguous requests to human agents. Monitor chat logs to identify patterns requiring improved knowledge base content or rule refinement.
</Info>

### Best practices

<AccordionGroup>
  <Accordion title="Review knowledge bases" defaultOpen>
    Audit connected knowledge bases to ensure they don't contain sensitive information you don't want AI to access. Remove outdated or incorrect documentation that could lead to wrong answers. Keep knowledge base content current and accurate.
  </Accordion>

  <Accordion title="Set clear escalation rules">
    Configure agents to escalate sensitive requests (HR issues, security incidents, legal questions) to humans. Define escalation criteria in agent instructions. Test escalation behavior before full deployment.
  </Accordion>

  <Accordion title="Monitor AI responses">
    Review conversation logs regularly to assess response quality. Track negative feedback patterns to identify knowledge gaps. Use <Tooltip headline="Agent results" tip="Categorization of conversation outcomes" cta="Learn about agent results" href="/documentation/automate/agents/deploy-monitor#agent-results">agent results</Tooltip> to measure performance and identify improvement opportunities.
  </Accordion>

  <Accordion title="Start with limited scope">
    Deploy agents to small channels or pilot groups initially. Expand gradually as you validate response quality and refine configuration. Test thoroughly before enabling auto-respond features.
  </Accordion>

  <Accordion title="Educate users">
    Train your team on AI capabilities and limitations. Set appropriate expectations about what agents can and cannot do. Provide clear guidance on when to request human assistance.
  </Accordion>

  <Accordion title="Maintain knowledge bases">
    Keep documentation current and comprehensive. Add articles addressing questions that generate "Not found response" results. Update content when product features or policies change.
  </Accordion>
</AccordionGroup>

***

## Support & compliance

### Contact information

For questions about AI trust, privacy, or data processing:

* **Privacy inquiries**: [privacy@ravenna.ai](mailto:privacy@ravenna.ai)
* **Security questions**: Contact your account manager
* **Data processing agreements**: Request through your account manager
* **Privacy policy**: [ravenna.ai/privacy](https://ravenna.ai/privacy)

### Compliance resources

* **SOC 2 Type II report**: Available upon request through your account manager
* **Data Processing Addendum (DPA)**: Contact your account manager
* **Subprocessor list**: Available upon request through your account manager
* **GDPR compliance documentation**: Contact [privacy@ravenna.ai](mailto:privacy@ravenna.ai)

### Data subject requests

Users can exercise data privacy rights:

* **Access**: Request copy of data processed by AI features
* **Deletion**: Request removal of conversation logs and AI-processed data
* **Correction**: Request updates to inaccurate information
* **Restriction**: Request limits on AI data processing

Submit data subject requests to [privacy@ravenna.ai](mailto:privacy@ravenna.ai) with required identification and specifics about the request.