PasteGuard

PasteGuard is a privacy proxy for teams that use AI with customer records, client files or project secrets. It runs locally or on your own server and checks requests before they reach a model provider.

• Data masking: Replace detected names, emails, phone numbers, financial identifiers and secrets with placeholders, then restore the originals in supported responses.
• Local model routing: Send requests containing sensitive data to a local LLM through Ollama, vLLM or llama.cpp, while requests without sensitive data can go to a configured cloud provider.
• Apps and coding agents: Protect requests through OpenAI-compatible and Anthropic APIs, including context sent by Codex, Claude Code, Cursor and Windsurf.
• Multilingual detection: Detect personal data across languages and scan for API keys, passwords, private keys, JWTs and connection strings, including in streaming responses.
• Request review: Use the local dashboard to inspect logged requests and see which values the proxy detected, masked and sent onward.

In Mask Mode, the cloud provider still receives the surrounding prompt and context with detected sensitive values replaced. Response restoration applies only where supported. Detection combines format and checksum checks with GLiNER for names and places.

PasteGuard is open source under Apache 2.0, supports Docker deployment and has no telemetry. Its privacy layer runs in your environment; requests forwarded to cloud providers still leave that environment. It can support stricter data policies, but it doesn't establish legal or security compliance.