SDSA
github.com
- Linux
- Web
- API
- Self-hosted
What SDSA is
SDSA is a free, open-source application for sanitizing tabular data before it leaves a trusted environment. It can be self-hosted on premises and used through a browser interface, REST API, or headless command-line pipeline for CI/CD and data pipelines. Inputs include CSV, delimited TXT, single-table SQL INSERT dumps, JSON records, and XML records. SDSA infers a schema and identifies fields that may contain personal information, such as email, phone, card number, government ID, date of birth, name, and address. Users can mask, HMAC-hash, tokenize, redact, drop, bin numeric values, truncate dates or strings, or add bounded Laplace noise. The application enforces k-anonymity for selected quasi-identifiers and can measure or enforce l-diversity for sensitive cleartext attributes. It exports sanitized CSV files with JSON and Markdown privacy reports containing utility summaries before and after processing. Uploaded data and outputs remain in an in-memory session store for a default 30 minutes. SDSA is licensed under AGPL-3.0.
Who it suits
SDSA suits teams handling enterprise data-sharing for testing, QA, analytics, or external processing. It is intended for organizations with explicit policies and reviewable privacy reports.
Good to know, and what to watch
- Accepts several common tabular and record formats
- Offers browser, REST API, and command-line workflows
- Supports multiple masking and redaction transformations
- Produces JSON and Markdown privacy reports
- Can enforce k-anonymity and l-diversity controls
- Sessions use in-memory storage for 30 minutes by default
- Multiple workers can break upload-to-download flow
- Does not claim dataset-level differential privacy
- Auxiliary-data linkage attacks may still succeed
The verdict
SDSA provides self-hosted sanitization with configurable transformations, privacy controls, and reports. Deployment needs careful handling of its single-process session design and stated privacy limits.
What the maker publishes
- Data handling
- Uploaded data and generated outputs live in an in-memory session store with a 30-minute default lifetime. source
- Data handling
- Uploaded data, parsed dataframes, HMAC keys, and generated outputs stay in process memory and expire after the configured session TTL, which defaults to 30 minutes. source
- Deployment
- The maker documents installation as a Python package or Docker deployment, including a production Docker Compose profile with nginx TLS termination. source
- Deployment constraint
- Deployments should run one SDSA application process unless the in-memory session store is replaced with shared storage, because multiple workers or replicas break the upload-to-download flow. source
- Deployment limit
- The maker says to run one SDSA application process per deployment while sessions use in-memory storage, because multiple workers or replicas can break the upload-to-download flow. source
- Deployment security
- The production Compose profile uses nginx for TLS termination, upload limits, security headers, and basic rate limits. source
- Input formats
- It accepts CSV, delimited TXT, single-table SQL INSERT dumps, JSON records, and XML records through its browser UI or REST API. source
- Intended users
- The maker describes SDSA for enterprise data-sharing workflows such as testing, QA, analytics, and external processing, with explicit policies and reviewable reports. source
- Interfaces
- Users can process data through a browser UI, REST API, or headless command-line pipeline for CI/CD and data pipelines. source
- License
- SDSA is licensed under AGPL-3.0. source
- License
- The repository states SDSA is licensed under AGPL-3.0. source
- PII detection
- It infers schema and detects likely sensitive fields such as email, phone, card number, government ID, date of birth, name, and address. source
- Privacy controls
- SDSA enforces k-anonymity over selected quasi-identifiers and can measure or enforce l-diversity for sensitive cleartext attributes. source
- Privacy controls
- SDSA enforces k-anonymity over operator-selected quasi-identifiers and can measure or enforce l-diversity for sensitive cleartext attributes. source
- Privacy limitation
- SDSA does not claim dataset-level (epsilon, delta) differential privacy and says auxiliary-data linkage attacks may still succeed. source
- Purpose
- SDSA is a self-hostable web app for sanitizing tabular data before it leaves a trusted environment. source
- Reports
- It exports sanitized CSV data with JSON and Markdown privacy reports, including before-and-after utility summaries. source
- Reports
- Each processed release includes sanitized data and JSON and Markdown privacy reports with before-and-after utility summaries. source
- Security limits
- SDSA says it does not provide dataset-level differential privacy or guarantee that auxiliary-data linkage attacks are impossible. source
- Supported data
- It accepts CSV, delimited TXT, single-table SQL INSERT dumps, JSON records, and XML records. source
- Transformations
- Available transformations include masking, HMAC hashing, tokenization, redaction, dropping, numeric binning, date or string truncation, and bounded Laplace noise. source
- Transformations
- Available field actions include masking, HMAC hashing, tokenization, redaction, dropping, numeric binning, date or string truncation, and bounded Laplace noise. source
- What it does
- SDSA is a self-hostable web app for sanitizing tabular data before it leaves a trusted environment, with privacy reports. source
- Workflow integrations
- SDSA provides a REST API and a headless command-line pipeline for CI/CD and data pipelines. source
Checked against the maker's pages on 9 Oct 2026.