Embed the privacy engine
datafog core
1.7.0 ↗Released
Local PII and credential detection and privacy transformations, with SDKs for Rust, Python, Node.js, and browsers.
In this release Restores email and phone detections missed in Core 1.5 and 1.6.
Protect sensitive data in your applications, check files before your AI agent reads them, and build encrypted local storage.
Choose datafog core, datafog MCP, fogvault, or the datafog Python library. Each product has its own release track.
python -m pip install datafog-core
Install the Core SDK in your application. Installation guide ↗
From application code to agent workflows and encrypted storage. Pick the product you need and follow its release track.
Release snapshot ·
Embed the privacy engine
Released
Local PII and credential detection and privacy transformations, with SDKs for Rust, Python, Node.js, and browsers.
In this release Restores email and phone detections missed in Core 1.5 and 1.6.
Protect files before your agent reads them
Released
Scan local files and create redacted, masked, or cleaned copies. Your agent receives counts and paths rather than matched values.
In this release Core 1.7 detection fixes, a reusable Python scanning interface, and vault-directory exclusions.
Build encrypted local storage
Released
An independently versioned Rust vault and browser SDK for encrypted persistence, locking, backups, and recovery.
Initial release Native SQLite storage and browser Worker/IndexedDB adapters, with encrypted backup transport.
Use the Python library and workflows
Released
The Python library, CLI, and workflow integrations, including a LiteLLM guardrail. Versioned separately from the Core SDK.
In this release Structured detection across engines, plus improved nested tool-argument and streamed-response handling in LiteLLM.
Coming soon
Local protection in your browser. The Chrome extension processes scanned content on your device.
Different workflows need different protections. Select the information to transform, apply rules by entity type, and exempt approved values.
Redact · Mask · Remove
Replace sensitive values with typed placeholders, mask them with optional partial visibility, or remove them from the text.
Explore transformations ↗Pseudonymize
Use keyed pseudonymization when repeated values need consistent replacements. Your application supplies the key provider.
Understand pseudonymization ↗Tokenize · Restore
Use provider-backed tokenization for reversible replacements. Your provider owns storage and authorization; Core checks the request scope.
See tokenization and restoration ↗Redaction, masking, and removal work across all SDKs. Pseudonymization and tokenization use application-supplied providers in Rust, Python, and Node.js.
Scan for findings, transform selected values, or combine both in one operation. The same Core defines the privacy behavior across SDKs.
Installation guide ↗Already using the datafog Python package? Core has a separate package and API. Follow the migration guide.
# pip install datafog-core
from datafog_core import scan_and_transform
result = scan_and_transform(
"Contact ada@example.invalid",
{"transform": {"default": {"strategy": "redact"}}},
)
print(result.text)
# Contact [EMAIL]
Explore agent plugins, gateway adapters, and other ways to use datafog. Each has its own setup and availability.
In development · 0.2.0-dev
A plugin for checking sensitive data in agent tool calls. For a released local file workflow, start with datafog MCP.
Follow plugin development ↗Python package adapter
An in-process guardrail adapter for the LiteLLM gateway.
Adapter setup and examples ↗npm release · 0.3.0
The fogclaw plugin brings PII detection and redaction to OpenClaw.
Explore fogclaw ↗datafog API ↗ · Docker distribution for self-hosted HTTP access. datafog Instructor 0.1.0b8 ↗ · An older beta for LLM-based PII extraction; requires an external model API.
Core provides the detection engine, MCP brings it to local agent workflows, Python adds application integrations, and fogvault provides encrypted persistence. Follow the product you use for its setup, release notes, and supported capabilities.
Available in Core today
Follow each product
See the product release highlights ↑ · Follow the Core roadmap ↗
Performance varies by workload. See the 0.3.0 benchmark methodology and results.
We fine-tuned a seven-category PII detector on a Mac and evaluated it on 90,018 held-out synthetic documents. Read the results, what regex still does better, and what remains unproven.
Read the research preview ↗Embed the engine, connect your agent, or build encrypted storage. Choose the datafog product that fits.