Keep sensitive data out of AI workflows.
Website · Documentation · MCP server · Core engine
DataFog builds open-source tools for local PII detection and privacy transformations.
- Use it with Claude Code: DataFog MCP scans local files and creates redacted, masked, or cleaned copies. The original stays unchanged. Tool responses report counts, locations, and copy paths without returning file contents or matched values.
- Build it into your application: DataFog Core provides the Rust engine and SDKs for Python, Node.js, and browsers.
Have a customer export with emails and phone numbers? Ask Claude Code to check it, then create a copy with those fields redacted before analysis. Detection and transformation run locally.
Install DataFog MCP · Try the sample CSV · Privacy boundaries
| Repository | Role |
|---|---|
| datafog-mcp | Local file scanning and redacted copies for AI agents. Stable v1.0.0, tested with Claude Code. |
| datafog-core | Recommended starting point for new applications. Shared detection and transformation engine, with Rust, Python, Node.js, and browser SDKs. |
| datafog-python | Established Python package, including the Claude Code hook and LiteLLM guardrail adapter. Moving to Core? Follow the migration guide. |
| datafog-claude-plugin | Claude Code integration, with its own setup and release path. |
| fogclaw | OpenClaw integration for PII detection and redaction. |
Core is a separate package and API from the established datafog Python
library. Existing integrations have their own capabilities and release paths;
Core adoption is a separate next step.
- Detect: scan text and parsed JSON records, including supported person-name fields, with local built-in detectors.
- Transform: redact, mask, remove, pseudonymize, or tokenize selected values. Select entity types, apply per-entity rules, and exempt approved values.
- Restore: restore provider-issued tokens under an exact request scope. Your provider owns storage and authorization.
Browser SDKs support detection and stateless transformations. Provider-backed pseudonymization, tokenization, and restoration are available in Rust, Python, and Node.js. See the SDK installation guide and transformation guide.
Core supplies detection and transformation. Your application or integration chooses where to invoke it and how to enforce policy.
One shared privacy engine across more application, agent, and gateway workflows. Core owns detection and transformation; integrations connect that behavior to the places data moves, including workflow-specific interception and enforcement.
Follow the Core roadmap for implementation status and the Core releases for shipped capabilities.
Website · Documentation · Installation · Contributing · Discord