Privacy Policy
Last Updated: April 11, 2026
Private by Design
Defrag is designed to keep your interactions private and under your control. We do not sell personal data. Any interaction content you add to your workspace stays private to your account and is not used to train public models.
What we collect
When you create an account and use Defrag, we collect account information (email, name), baseline signals you choose to provide, and the interaction content you add to the workspace. We also collect minimal operational data necessary to provide and secure the service (logs, usage metrics, error reports).
- Account identifiers: email address, display name
- Workspace content: sessions, reads, notes you create
- Baseline signals: optional data you provide about your context
- Usage and error logs for service reliability
How we use data
We use your data to provide the service: generating reads, preserving continuity across moments, and helping you track patterns over time. We may use aggregated, de-identified metrics to improve the product, but we do not use your identifiable interactive content to train public models.
Retention Timeline
- By default, your workspace interactions are retained to enable continuity and pattern recognition.
- You may request deletion of specific interactions or your entire account at any time via Settings.
- After a verified deletion request, associated content is removed within 30 days, subject to legal retention requirements.
- Inactive accounts may be purged after 12 months of inactivity following advance notice.
Deletion Expectations
- You can access, correct, export, or request deletion of your data from the Settings panel.
- Deletion requests are processed promptly and completed within 30 days of verification.
- Some data may be retained as required by law or for security purposes, but will not be used for any other purpose.
Contact
For privacy questions, data requests, or concerns, contact us at info@defrag.app. We are a small team and respond personally.
Model usage and AI
Defrag uses models to power reads and explainers. We do not expose your identifiable interaction content to public model training. Models may be updated and improved using aggregated, de-identified signals.