Pasteguard

PasteGuard fills a narrow but essential niche: letting regulated teams use cloud AI without manual redaction. The auto-detection of over 30 PII types works reliably across languages, and the placeholder system means the AI still understands context. We'd reach for this when we need to paste client data into ChatGPT or send production .env files into coding agents. The local-first architecture means no third-party relay—data only goes to the AI provider after masking. That's a strong privacy posture compared to cloud-based data loss prevention services. The open-source core and Apache 2.0 license are good for audits. Where it bites: PasteGuard isn't a fully offline solution. Route mode can keep some sensitive requests on a local LLM, but most traffic still hits cloud endpoints after masking. If your compliance requires zero data to leave your network, you'll need a fully local model. Also, the browser extension is still experimental—expect bugs. The Docker setup is simple, but non-technical users may struggle. Compared to alternatives like PrivateGPT or local-only models, PasteGuard trades full privacy for AI capability. You get top-tier models (GPT-5, Claude) with masked data instead of a weaker local model. For most compliance scenarios, that's the right trade-off. In practice, teams adopt it for three reasons: (1) it works with any provider, (2) it requires no changes to existing AI workflows, and (3) it's free and self-hostable. The live dashboard is a nice bonus for monitoring. Bottom line: If you're a developer or compliance officer who needs cloud AI without data exposure, PasteGuard is the most pragmatic tool out there. Just account for the technical setup overhead and the fact that it's not an air gap.