Frequently Asked Questions

Custos is a privacy-first accountability software that helps you stay focused and avoid harmful content online. It uses on-device AI to monitor your screen and intervenes when it detects concerning content, engaging you in a reflective dialogue rather than just blocking access.

Unlike traditional blockers that simply block websites or keywords, Custos:

• Uses AI to understand visual content, not just URLs
• Runs entirely on your device for complete privacy
• Engages you in dialogue rather than just blocking
• Cannot be easily bypassed like browser extensions
• Works across all applications, not just browsers

Custos currently supports macOS, Windows, and Linux. Mobile support is on our roadmap but presents unique challenges due to platform restrictions.

No, never. All image analysis happens locally on your device using the CLIP AI model. Your screenshots are processed in memory and immediately discarded. They are never saved to disk or transmitted anywhere.

Self-hosted (free): Zero data collection. Everything stays on your machine.

Patron tier with cloud sync: Only minimal metadata is synced: timestamps of interventions, session IDs, and unlock events. The actual content that triggered an intervention is never transmitted.

No. Custos is designed for personal accountability, not surveillance. There is no "accountability partner" feature that shares your activity. Your data stays between you and your Guardian AI.

If you cancel your Patron subscription, your cloud-synced audit logs are deleted within 30 days. Your local installation continues to work as a self-hosted setup, with all data remaining on your device.

The CLIP AI model analyzes your screen for visual patterns associated with harmful content. This includes explicit imagery, provocative content, and related visual elements. The detection is based on visual similarity, not keyword matching.

Yes, you can adjust the sensitivity threshold in your configuration. A lower threshold (e.g., 0.20) is more sensitive and may produce more false positives. A higher threshold (e.g., 0.30) is less sensitive. The default (0.25) works well for most users.

You can temporarily disable Custos through the system tray icon before accessing legitimate content that might trigger a false positive. We recommend setting a timer to remind yourself to re-enable it. The Guardian trusts your judgment for legitimate use cases.

You can exit anytime—the intervention is a voluntary moment of reflection, not a lockout.

Three ways to exit:

• Click the Exit button at the top of the overlay
• Press the Escape key
• Ask the Guardian to let you go (e.g., "let me leave")

The Guardian is a compassionate companion, not a prison warden.

Custos is designed to be lightweight. It captures a screenshot every few seconds and processes it using an optimized CLIP model. Most users don't notice any performance impact. If you have an older machine, you can increase the scan interval in the configuration.

$9/month with a 7-day free trial. Cancel anytime. This includes everything: one-click installer, Guardian AI, journey tracking, and priority support.

Yes! 7 days free, no credit card required to start. Try Custos risk-free and cancel anytime if it's not for you.

Custos is open source, so developers can self-host. However, this requires Python, Docker, terminal commands, and your own OpenAI API key (~$1-5/month).

The self-hosted option is for developers who want to inspect the code or have specific infrastructure requirements. For everyone else, the $9/month subscription is the recommended experience.

We use Stripe for secure payment processing. You can pay with any major credit card, debit card, or Apple Pay/Google Pay.

CLIP (Contrastive Language-Image Pre-training) is an AI model developed by OpenAI that understands both images and text. Custos uses a local version of CLIP to analyze screenshots and determine if they contain concerning content, without sending images anywhere.

On first run, Custos downloads the CLIP model (~500MB). This is a one-time download. Subsequent launches are much faster as the model is cached locally.

Yes. Custos monitors all connected displays and can detect concerning content on any screen.

The server component can run on any machine with Docker. However, the desktop client needs to run on the machine you want to monitor, as it captures the screen.