Security and data protection

Odella uses a layered security model for identity, customer separation, worker execution, connected applications, model access, and human approval. Product controls and customer configuration work together: Odella operates the platform controls, while each customer decides who may access its account and what each worker is permitted to use.

​

Security model

​

Identity and least privilege

Access decisions can involve several layers:

1. The authenticated organization and user
2. The selected AI Employee
3. Resources assigned to that worker
4. The current conversation, workflow, or task
5. Tool-specific allow, block, and approval rules
6. The connected account’s own permissions

A valid user session or active worker does not grant access to every organization resource. Assign workers only the capabilities required for their responsibilities and remove access when it is no longer needed. See Security & Privacy settings for inbound access, outbound messaging, tool approval policies, and saved exceptions.

​

Data protection

The technical and organizational measures described in the DPA include, as applicable:

• Role-based and least-privilege access
• Unique identities and multi-factor authentication for privileged and critical production access
• Access logging and periodic access review
• Encryption in transit using TLS 1.2 or higher
• Encryption at rest using AES-256 or equivalent provider-managed encryption
• Managed secret storage and access restrictions
• Logical customer-separation controls
• Development and production separation
• Change review, testing, deployment records, and recovery practices
• Vulnerability, dependency, and cloud-security monitoring
• Security logging, incident response, backup, and recovery procedures
• Vendor and Subprocessor review
• Documented deletion and disposal procedures

The complete and current contractual description is in Annex II of the DPA.

​

AI Employee execution

Worker execution is separated from the public application tier. Persistent customer work and replaceable execution capacity are treated separately, and tool availability is resolved from the worker’s authorized resources and policies. Important customer controls include:

• Worker role and manager
• Assigned skills, workflows, templates, and knowledge
• Connected inboxes and applications
• Working hours and service state
• Inbound and outbound messaging policies
• Tool approval policy and saved tool rules
• Human approval steps inside workflows

See AI Employee settings for the complete configuration reference.

​

Integrations and credentials

Odella supports OAuth and other appropriate authentication methods for customer-authorized applications. Connection and authorization are completed through controlled application flows, and protected credentials are associated with the relevant account or user scope. Workers normally use connected applications through authorized tools. They do not need reusable application credentials in their workspace. A connected application remains subject to:

• The provider’s permissions and terms
• Organization and user scope
• Worker assignment
• Tool approval rules
• Outbound-recipient policy
• Workflow approval steps

Custom integrations connect customer-selected remote MCP servers and carry additional third-party risk. See Custom integrations for endpoint requirements, authentication, assignment, and the conservative tool trust model.

​

Model access

Model requests pass through a controlled Odella service boundary. Worker environments are not provisioned with direct model-provider credentials. Available models and worker model selections remain subject to workspace and worker configuration. See Models and Model & Intelligence settings.

​

Data residency and international processing

Data residency, international transfers, support access, customer-selected services, and Subprocessor locations are governed by the applicable contractual terms and account configuration. Do not assume that choosing a region prevents every transfer. Customer-selected model providers and integrations may process data in locations governed by their own terms and configuration. Review:

• Privacy Policy for current privacy, residency, retention, and individual-rights information
• Data Processing Addendum for processing roles, security terms, international-transfer mechanisms, breach obligations, and deletion terms
• Subprocessors for how the current list and change-notice process are managed
• Trust Center for current assurance and Subprocessor information

​

Monitoring, audit, and resilience

Odella records relevant operational and security events to support monitoring, troubleshooting, access review, and incident response. Platform controls include protected logging, monitored services, bounded retries, recovery mechanisms, and separation of durable customer state from replaceable worker execution. The product’s Activity Log gives authorized users a worker-level view of conversations, tasks, workflows, and service events. It is not a replacement for the broader assurance material available through the Trust Center.

​

Shared responsibility

​

Recommended customer controls

​

Assurance and legal resources

For privacy and Data Subject requests, contact dsar@odella.ai. For legal and DPA questions, contact legal@odella.ai.