Modules
Obscura is a library-first privacy toolkit for Elixir.
String analyzer facade for Obscura recognizers.
Basic recognizer explanation for deterministic decisions.
Normalized analyzer result.
String anonymizer facade for validated operators.
Structured, value-safe anonymizer operator error.
Metadata for one anonymization replacement.
Result of an anonymization run.
Machine-readable optional dependency and model asset contracts.
Structured, report-safe diagnostic returned by optional runtime setup paths.
Provider-independent LLM message redaction and rehydration helpers.
Safe language normalization for analyzer options.
Behaviour for optional language detectors.
Logger-safe helpers for redacting terms before applications log them.
Dependency-light NLP artifacts for analyzer scoring.
Behaviour and dispatcher for analyzer-level NLP artifact engines.
Optional analyzer-level NLP engine for Bumblebee token-classification serving.
Behaviour and guarded runtime for application-defined anonymizer operators.
Versioned salted hashing for anonymizer replacements.
Plug-compatible request sanitization helper.
Validation helpers for local native privacy-filter checkpoints.
Development setup helper for local native privacy-filter checkpoints.
Reusable native privacy-filter serving runtime.
Product profiles and runtime requirement checks.
Supervised asynchronous preparation and ownership of a reusable profile runtime.
Explicit reusable runtime resources for product profiles.
Behaviour for Obscura recognizers.
Optional GLiNER recognizer facade.
Experimental native Nx/Emily serving for the pinned Urchade GLiNER model.
Optional Ortex-backed serving for GLiNER span models.
Explicit NER recognizer for open-class entities.
Deterministic fake NER serving used by fixtures and tests.
Normalized configuration for an optional local NER model.
Optional Nx/Bumblebee serving builder for local token-classification models.
Data-backed custom pattern recognizer definition.
Optional ex_phone_number validator for parser-backed phone recognition.
Optional native privacy-filter recognizer.
Protocol for application data that can define its own Obscura redaction policy.
Rehydrates vault tokens in strings.
Structured data rehydration for vault tokens.
Streaming-safe vault token rehydration.
Public structured data redaction API.
Redaction metadata for one structured data leaf.
Result of a structured redaction run.
Tiktoken-compatible byte-pair encodings used by Obscura.
A tiktoken-compatible byte-pair encoding.
Facade for reversible pseudonymization vault backends.
Behaviour for optional vault backends.
Session-scoped ETS-backed vault.
One reversible pseudonymization mapping.
Session-scoped in-memory vault.
Mix Tasks
Detects PII in a file or stdin using Obscura.
Verifies local links in repository Markdown files.
Runs opt-in real local NER model smoke validation.
Runs authoritative operational benchmarks for measured product profiles.
Promotes one complete operational report into its separate manifest.
Runs one long-duration operational soak.
Promotes one complete soak report into the separate soak manifest.
Verifies the authoritative soak manifest and promoted artifact hashes.
Verifies operational manifest structure and promoted report hashes.
Validates a local native privacy-filter checkpoint.
Downloads a local native privacy-filter checkpoint and validates it.
Checks whether an Obscura product profile is ready.
Explicitly prepares and validates a reusable Obscura profile runtime.
Redacts PII in a file or stdin using Obscura.