PasteGuard

PasteGuard is an open-source privacy layer designed to mask sensitive information before it is sent to AI models such as ChatGPT, Claude, Gemini, and coding agents. It addresses the challenge of protecting private data—including names, emails, secrets, customer details, and a wide range of credentials—when interacting with AI providers or integrating AI into applications. By replacing sensitive elements with placeholders, PasteGuard ensures that the context required for AI processing is preserved while the original data remains confidential and is not exposed to external providers.

The tool detects and masks over 30 types of private data and secrets in any language, including names, emails, phone numbers, credit card numbers, IBANs, EU VAT numbers, IP addresses, locations, API keys, SSH keys, JWT tokens, passwords, connection strings, private keys, and bearer tokens. PasteGuard operates by intercepting user input before it reaches the AI provider, substituting sensitive information with standardized placeholders (such as [[PERSON_1]], [[EMAIL_1]], or [[API_KEY_1]]). Where supported, it can also restore the original values in AI responses for the user’s convenience.

PasteGuard is suitable for teams and organizations with strict privacy requirements, such as those needing to comply with GDPR, DORA, client confidentiality agreements, or internal AI usage policies. It is particularly useful for scenarios where raw customer data, client files, or secrets cannot be shared with AI services. The platform can be used locally, self-hosted in a team’s own environment, or integrated in front of apps and SDKs to mask data before it reaches APIs from providers like OpenAI and Anthropic.

Delivery options include a browser extension—currently in beta—for ChatGPT, Claude, and Gemini, as well as Docker-based self-hosting for integration with OpenAI-compatible apps, SDKs, and coding agents. PasteGuard is fully open-source, auditable, and does not include telemetry or tracking. 0 license, allowing teams to deploy and audit the tool within their own compliance boundaries.