Composable, multimodal toolkit for detecting and redacting sensitive data.
Find and remove PII and PHI across text, images, audio, and tabular data, with a full audit trail for every entity.
Elide gives you the pieces (recognizers, deduplication, validation, redaction operators, and format codecs) and you wire them into your own document flow. It is the toolkit layer only: the orchestrating runtime and the gateway server live in separate projects.
Warning
Active development: API not stable. This project is under active development. Public APIs, configuration shapes, and on-disk formats may change without notice between releases. Pin a specific commit if you depend on this in production.
Multimodal
One entity model spanning text, images, audio, and tabular data, so a recognizer or operator written once serves every format.
Detection
Regex, dictionary, and checksum recognizers with validators, plus NER and LLM/VLM recognition. Language, OCR, and speech enrichers feed the text they produce back into the same pipeline.
Context-aware scoring
Nearby keywords lift ambiguous matches, and overlapping findings reconcile into one deduplicated entity set.
Redaction operators
Mask, replace, truncate, HMAC, hash, generalize, or clamp text; blur, pixelate, or black out image regions; silence or beep audio; drop rows and columns. Reversible encryption and pseudonymization round-trip.
Format codecs
Read and rewrite TXT, CSV, JSON, XML, HTML, RTF, PDF, DOCX, PPTX, XLSX, images (PNG, JPEG, TIFF), and audio (WAV, MP3), changing only the redacted spans and leaving every other byte intact.
Provenance-first
Every entity carries its full audit trail of how it was found, scored, and hidden, and the trail verifies.
Everything is feature-gated: take only the modalities, recognizers, and codecs you need.
A live demo runs the pattern-and-dictionary
detection and redaction pipeline compiled to WebAssembly, entirely in your
browser: paste text, redact it, and nothing you type leaves the page. The source
is crates/elide-wasm; build it locally with make wasm-demo.
docs/ develops the conceptual architecture, one document per stage
of a document's path through the toolkit.
| Document | Subject |
|---|---|
| Ingestion | Raw bytes to a typed, addressable handle: format resolution, streaming, and coordinate promotion. |
| Detection | Composing rule-based, statistical, and generative recognizers into one annotation set. |
| Redaction | Turning entities into concrete rewrites: operator kinds, leak profiles, and the reversible boundary. |
API documentation lives at docs.nvisy.com.
- Changelog: CHANGELOG.md for release notes and version history
- Contributing: CONTRIBUTING.md
- License: Apache 2.0, see LICENSE.txt
- Documentation: docs.nvisy.com
- Email: support@nvisy.com