# Data & Security

> What Modeless does with your data, in plain language: data flow, what is stored, retention, logs, credentials.

Source: https://modeless-web.vercel.app/security

The plain-language version for platform teams and security reviewers: what Modeless does with your data, not the legal commitments. Those are in the [Privacy Policy](https://modeless-web.vercel.app/privacy).

- **Data flow**: Your client → Modeless → the model providers that serve the request. Auto additionally sends the task and the agent's working history to a model that judges task difficulty: once per agent session, or every few turns of a plain chat until it upgrades.
- **Stored**: Per-session routing state (which model the session is on, and whether it has been assessed). For Fusion, the panel's answers for the current turn, so tool calls in that turn reuse them.
- **Retention**: Routing state and the Fusion panel cache expire automatically. (TTLs TBD)
- **Logs**: Request metadata: timing, status, token counts, which model answered. Responses, tool content and API keys are not logged. The text of each new user message is logged, with secrets masked and capped at 5,000 characters. (confirm before launch: on in current production config)
- **Credentials**: API keys are never written to logs or to the routing store.
- **Region**: (processing region TBD)
- **Certifications**: (none yet; listed here when obtained)

Found a vulnerability? Write to security@modeless.ai.