Dataiku Releases Kiji Privacy Proxy: Open-Source Privacy Layer for AI APIs
Key Takeaways
- ▸Detects and masks 26 types of PII automatically using ML-powered detection before data is sent to AI APIs
- ▸Runs entirely locally with ONNX-optimized inference—no external API calls or data exposure
- ▸Works as a transparent proxy requiring zero code changes, with automatic browser configuration on macOS
Summary
Dataiku's 575 Lab has released Kiji Privacy Proxy, an open-source tool that automatically detects and masks personally identifiable information (PII) in requests to AI services like OpenAI and Anthropic. The tool uses ML-powered detection to identify 26 different types of PII—including emails, social security numbers, credit cards, and addresses—and replaces them with realistic dummy values before sending data to external APIs, ensuring sensitive data remains local and protected.
Kiji operates as a transparent proxy that requires zero code changes and works seamlessly across multiple platforms. For macOS, it provides automatic proxy configuration (PAC) with native integration into Safari and Chrome, while Linux users can run it as a standalone server. The tool also offers a Chrome extension for inline PII detection on popular AI chat interfaces like ChatGPT, Claude, and Gemini. All processing happens locally using an ONNX-optimized DistilBERT model, with sub-100ms latency and no external API dependencies.
The release targets organizations that need to comply with privacy regulations (GDPR, HIPAA, CCPA) while leveraging AI services, as well as development teams seeking to prevent accidental data leaks. Kiji is production-ready with systemd service support, Docker compatibility, and comprehensive logging, available for download on GitHub.
- Available for macOS and Linux with Chrome extension support for ChatGPT, Claude, and other AI chat interfaces
- Helps organizations comply with GDPR, HIPAA, CCPA and prevent accidental data leaks in AI workflows
