# Data Processing

The data Risotto processes, how it processes it, which model providers it uses, and where and how the data is stored and encrypted

## tl;dr

**What data is stored?**

- Organizational data (company structure, user profiles)
- Communication data (Slack and Microsoft Teams messages, files, channels)
- Knowledge base content from 12+ sources (Confluence, Jira, Google Drive, GitHub, etc.)
- Ticketing workflows
- Integration credentials
- Usage analytics

**How is data processed?**

The data above flows through Risotto's multi-tenant architecture and is processed by large language models from OpenAI and Anthropic, which Risotto reaches through the OpenAI API, the Anthropic API (including Claude Platform on AWS) and AWS Bedrock. We use these models to classify, extract, and understand text; generate vector embeddings for semantic search; orchestrate automated workflows; and keep systems in sync via webhooks—all with organization-specific isolation and comprehensive security controls.

*Product example: Diagram: requests, knowledge and identity data flow into the organization's workspace, which uses model providers and stores data in a primary database, a search database and file storage.*

## 1. What Data Will Be Processed by Risotto

Risotto processes organizational data including user profiles, department details and memberships, application and access rules, identity provider data (Okta, Microsoft Entra ID, Google Workspace, JumpCloud), and employee data from connected HR systems (BambooHR, HiBob, Rippling, Workday). Communication data from Slack includes messages, files, and channel configurations. Risotto also processes similar communication data from Microsoft Teams, email, and the Jira, ServiceNow and Freshservice request sources. Knowledge base content is ingested from Confluence, Notion, Google Drive, SharePoint, OneNote, Coda, GitHub, Jira, Zendesk, Freshservice, Guru, and Slack channels, including page content, metadata, and user permissions.

## 2. How Data Is Processed

Risotto's multi-tenant architecture enforces strict isolation between organizations with department-level access controls. AI-powered processing converts text into vector embeddings for semantic search, intelligently segments long documents while preserving context, and automatically extracts metadata from all sources. Using retrieved knowledge and conversation context, natural language processing enables accurate query understanding and response generation. Workflow automation then categorizes and routes requests, orchestrates parallel or sequential approvals, manages escalations, and sends real-time notifications. Data stays current through real-time webhooks and scheduled batch jobs with conflict resolution. Throughout, security controls apply AES-256 encryption with organization-specific keys, role-based access control, and comprehensive audit logging.

For how indexing and search work, see [Indexing Best Practices](https://www.tryrisotto.com/docs/concepts/indexing).

## 3. How Data is Stored

**Primary PostgreSQL Database**: A high-availability PostgreSQL cluster stores organizational data, including department details, user memberships, ticketing workflows, conversation histories, access request states, approval chains, integration configurations, and knowledge management metadata. Sensitive data such as integration credentials are encrypted with AES-256 using organization-specific keys before being stored; the database itself is encrypted at rest with the same algorithm, so this critical data is effectively encrypted twice.

**Vector Database (PGVector)**: A separate, high-availability PostgreSQL cluster stores high-dimensional text embeddings and keywords for hybrid semantic search, along with document metadata, source information, and department associations. Each organization's content is in its own table. This data is also encrypted at rest with AES-256.

**S3 File Storage**: Stores document attachments, media content, compressed exports, and backup archives. Access controls include buckets with public access blocked, region-specific deployment, and bucket-level restrictions that prevent public ACLs, public policies, and unauthorized access.

## Data Security and Compliance

Risotto protects customer data with AES-256 encryption and organization-specific keys, ensuring complete tenant isolation. Role-based access control with department-level permissions enforces segregation across the multi-tenant architecture. Vector embeddings are protected by database-level security controls and organization-specific table isolation. The [Audit Log](https://www.tryrisotto.com/docs/dashboard/settings/audit-log) records configuration changes in your organization and who made each one, and you can stream it to a SIEM with an API key. Tool runs and approvals on a ticket are recorded on the ticket. See [Ticket details](https://www.tryrisotto.com/docs/dashboard/tickets/ticket-details).

### Organization-specific encryption keys

Integration credentials, such as API tokens and client secrets, are encrypted with envelope encryption:

- Each organization has its own data key, an AES-256 key that Risotto generates in AWS Key Management Service (KMS).
- The data key is stored only in encrypted form. A master key that never leaves AWS KMS encrypts it, and Risotto asks KMS to decrypt it when it needs it.
- Risotto encrypts each credential with the organization's data key, using AES-GCM.
- Each organization's data key is replaced every 30 days, and its stored credentials are encrypted again with the new key.

*Product example: Diagram: a master key in AWS KMS wraps each organization's AES-256 data key, which encrypts that organization's credentials.*

## IP allowlisting

If your firewall or a connected system only accepts traffic from approved IP addresses, add Risotto's static IP addresses to your allowlist:

- `54.176.23.214`
- `54.151.22.32`

Source: https://www.tryrisotto.com/docs/security/overview
