Data Flow Description

This description explains in detail how customer content moves through and is processed on the Lyyli platform. It helps you understand the system's technical architecture, but it does not replace the precise processing regions of individual AI routes or the service-specific retention terms that apply to them.

Six-step information flow

The lifecycle and processing of customer-produced content in Lyyli follows this model:

• Content import or creation: A user writes new content or imports existing material into the Lyyli platform.
• Storage in the customer workspace: Content is stored securely in the workspace according to retention settings defined for that organisation.
• AI request preparation: For an AI call, Lyyli builds a technical request and seeks to minimise unnecessary identifying information where applicable.
• Transfer to Vercel AI Gateway: The request is sent via Vercel AI Gateway to the language model selected by the customer.
• Processing at the model provider: Anthropic (Claude, EU), Google Vertex AI (Gemini, EU), or OpenAI (GPT, Global) processes the request under Zero Data Retention. Processing region is written to logs and is auditable.
• Response storage: The language model's response returns to Lyyli. Lyyli stores the conversation in its own EU environment to deliver the service.

Logging practices

The system maintains logs of usage and actions that include the following information:

• User identifier, action performed, model or route used, processing region (EU or Global), precise timestamp, and success or error status.
• Metadata related to billing and resource tracking, such as token counts used.
• Whether input or response content is stored in system logs depends on workspace settings and the log type in question.

Separate data flows

Lyyli's security relies on a clear separation of independent data flows:

Workspace storage

Primary application data such as drafts, user data, and project data is stored securely in databases located in the EU region.

AI processing

Language-model requests go through Vercel AI Gateway. The model provider does not retain the prompt or response after the request (ZDR). Claude and Gemini inference is locked to the EU; GPT processing is global. Lyyli stores conversations in the EU workspace. Image generation, web search, and meetings are not covered by language-model ZDR.

Integrations

Data transfer to external systems occurs through integrations activated by the organisation. These include Microsoft Teams, Slack, email connections, calendars, and meeting platforms.

Security and audit logs

System security logs and audit trails are retained securely under a separate retention schedule, ensuring traceability of events even after possible workspace deletion.

Backups

Automated system backups rotate securely and in a controlled manner for a limited period, after which they are overwritten automatically.