Skip to main content

​
General

Nvisy is an open-source multimodal redaction platform. It detects and removes sensitive information across documents, images, and audio, combining deterministic patterns, NER, computer vision, and LLM-driven classification into auditable, policy-driven pipelines.
Yes, Nvisy is open-source. You can find our code on GitHub and contribute to the project.
Nvisy can detect:
  • PII: Names, addresses, SSN, phone numbers, email addresses
  • PHI: Medical records, diagnoses, prescriptions, patient data
  • Financial: Credit card numbers, bank accounts, tax IDs
  • Custom patterns: Your organization-specific sensitive data
Each finding carries a calibrated confidence score and a provenance chain showing how it was recognized. A detection stops after analysis so findings can be reviewed before anything is modified — detection informs a decision rather than replacing one.
Documents (pdf, docx, rtf, txt, log, html, xml, json), tabular data (csv, xlsx), images (png, jpg, tiff), and audio (wav).View complete list →

​
Pricing & Plans

Nvisy is in private beta. Contact us to discuss pricing and access.
Yes. Nvisy is open-source and can be self-hosted, which is the right choice for strict data-residency requirements.On-premise deployment →

​
Technical

Yes. Nvisy provides a REST API and official SDKs for TypeScript, Python, and Rust.View API documentation →
Retention is set per workspace and can be overridden per pipeline. The default for every scope is ephemeral — data is deleted as soon as it has been processed.A workspace sets retention for four scopes: original documents, redacted documents, enrichment intermediates (OCR layout, transcripts), and audit logs. Each is persistent (keep indefinitely), ephemeral, or fixed with a number of days.A pipeline may override retention for the scopes it produces: redacted documents, intermediates, and audit logs. Original documents are ingested rather than produced, so they always follow the workspace baseline.Audit trails →
Workspaces connect to object storage (s3, azure, gcs) and to cloud file services (google_drive, dropbox, one_drive, box) for moving documents in and out, and to LLM providers (openai, anthropic, ollama) for model-driven detection. Webhooks deliver events as detections and syncs progress.View all integrations →
Official SDKs:
  • TypeScript/JavaScript — full API coverage
  • Python — HTTP client; typed service methods are in progress
  • Rust — HTTP client; typed service methods are in progress
You can also use the REST API with any language.
Yes! Nvisy includes OCR (Optical Character Recognition) to extract and redact text from scanned documents and images.

​
Security

  • Workspace isolation as the tenancy and access-control boundary
  • Per-workspace retention, overridable per pipeline
  • A reviewable detection phase — nothing is modified until redaction is applied
  • Verifiable audit trails with per-entity provenance
  • Self-hosting, so data and inference stay inside your own infrastructure
Self-hosting puts data location under your control. Nvisy is open-source, so the platform itself runs inside your own infrastructure.Where documents actually reside still depends on how you configure it. A workspace can sync files to s3, azure, or gcs, and detection can call a hosted LLM provider — either will send content outside your network. Documents stay entirely within your infrastructure only when every configured storage target and integration is local or private, for example local storage paired with an ollama connection so inference runs in-network too.On-premise deployment →
With a self-hosted deployment, we have no access to your documents. They stay within the infrastructure you point Nvisy at — keep storage and model connections local or private if you need them to stay entirely in-network. Retention is yours to set per workspace and override per pipeline, including ephemeral to delete data as soon as it has been processed.
Yes! Every redaction includes a complete audit trail showing:
  • Who initiated the redaction
  • What was detected
  • What was redacted
  • When it occurred
  • Confidence scores
Learn about audit trails →

​
Getting Started

  1. Sign up at nvisy.com, or self-host
  2. Create a workspace and issue an API token
  3. Install an SDK or call the REST API
  4. Upload a file, run detection, review, then redact
View quickstart guide →
For the API and SDKs, yes. A web interface is also available for reviewing detections and approving redactions without writing code.
Yes. Nvisy is open-source, so you can self-host and evaluate the full platform locally. For cloud access during private beta, contact us.
We offer:

​
Features

Yes. A policy targets labels from the deployment’s taxonomy and names the operator to apply to each, per modality — for example blackbox, blur, or pixelate for images, and mask, replace, pseudonymize, or encrypt for text. Reviewers can override any individual detection before redaction is applied.Redaction workflow →
Yes! Nvisy supports batch processing with parallel execution for faster processing of multiple documents.
Yes, Nvisy preserves the original document structure, fonts, images, and layout while redacting sensitive information.
Yes. Register a webhook to receive events as detections and syncs progress, or subscribe to a detection’s server-sent events stream for live status.Learn about webhooks →
You can:
  1. Adjust the confidence threshold
  2. Add custom patterns for organization-specific data
  3. Review detections before finalizing
  4. Manually mark additional items for redaction
Yes. A policy targets specific labels from the deployment’s taxonomy and names the operator to apply to each. Query GET /capabilities/labels to see what a deployment supports, then reference those labels from your policy.Because keep is an operator, a policy can also record that something was recognized and deliberately left intact.

​
Deployment

  • Cloud: Fully managed, auto-scaling
  • On-premises: Complete control in your infrastructure
  • Hybrid: Combine cloud and on-premises
Compare deployment options →
Illustrative production starting point:
  • 16+ CPU cores
  • 32 GB+ RAM
  • 500 GB+ NVMe SSD
  • 10 Gbps network
  • Docker or Kubernetes
These are preliminary guidance rather than benchmarked minimums — size from a trial run with your own documents.View complete requirements →
Yes! Nvisy can be deployed on AWS, Azure, Google Cloud, or any Kubernetes-compatible environment.
Yes, enterprise customers can request professional services for deployment, training, and custom integrations.

​
Troubleshooting

Common causes:
  • API key not set in environment variables
  • Using test key in production (or vice versa)
  • API key expired or revoked
  • Extra spaces or characters in the key
Troubleshoot authentication →
Processing time depends on:
  • Document size (pages)
  • File format (PDFs are fastest)
  • Detection types enabled
  • Current system load
Typical documents finish in under 10 seconds, though this is not a committed figure — large documents, OCR-heavy scans, and busy instances take longer.
Retry with exponential backoff. In the TypeScript SDK, NvisyApiError.isRetryable() tells you whether an error is worth retrying — it returns true for 5xx, 408, and 429.Only retry automatically when the request is idempotent. Starting a detection accepts an Idempotency-Key header, so reusing one stable key across attempts returns the existing detection rather than starting a second one.
Check:
  • File size — a 413 means the upload exceeded the workspace’s maxUploadBytes cap, or the server-wide hard limit
  • File format is supported
  • File is not corrupted
  • Your token has upload permissions in that workspace
Uploads are atomic per batch: if any document in the request fails, none are recorded.

​
Still Have Questions?

Documentation

Read our comprehensive guides

GitHub

Report issues or contribute

Email Support

Contact our support team

Sales

Talk to sales about enterprise needs