Sensitive files leave your systems every day unprotected. Pipeline sits inline in the workflows already moving them: scan, apply policy, output a protected version before the file arrives.
Pipeline protects exactly the entities that need it. The SSN, the patient ID, the account number. Run it and watch the rest of the document stay fully readable.
Files move constantly: to courts, to partners, into AI systems. Most teams answer with DLP, which flags risky transfers but leaves the content untouched, or manual redaction, which is slow and error-prone. Either way, no protection travels with the file. Once it leaves your environment, the sensitive data inside is readable by whoever receives it.
A blocked send, a warning, or a manual redaction job routed to someone with other priorities. The file stays in plaintext, exactly as exposed as the day it was created.
The file enters Pipeline, is scanned and protected, and is output before it reaches its destination. No ticket. No manual step. No removable overlay.
Every file follows the same path whether it came from a drag-and-drop, a folder watch, or an application API call.
Via API call, folder watch, or app event; one file, a batch, or a continuous stream. No upstream changes required.
NLP-based detection identifies PII, PHI, financial, and custom entities, each returned with a confidence score.
Chose whether it's encrypted, redacted, or tokenized. Different data handled differently in a single pass.
Placed in a folder, a destination app, or the next stage of an AI pipeline. Format preserved, content protected.
Protection becomes available at the entity level, not only the file level. Applied the moment a file moves, so the document stays usable and the sensitive content never travels unprotected.
Most tools work after the fact with overnight scans and bulk jobs. Pipeline fires the instant a file transitions. Created at 3:00 and emailed at 3:01? It's protected before it sends.
Full-file encryption makes documents unusable. Pipeline touches only what policy specifies, such as the SSN in field 4, the salary in row 12. The rest stays readable and workflows aren't broken.
No rebuilds, no new tools. Anything that can fire an HTTP call can integrate. Deploys in your cloud or on-prem: Docker, Kubernetes, ECS, or EKS.
Encryption keeps content accessible to authorized users. Redaction removes values permanently, which means they are court-ready. Tokenization swaps entities for length-matched hashes, recoverable for AI pipelines.
Configurable per profile, from surgical entity-level up to full-document protection on any match. Conditional logic on metadata, file type, or existing Microsoft Purview labels.
Key server, analysis engine, and AI Guard workers each scale independently. Batch mode runs in parallel; streaming mode processes large files in chunks, so there is no wait for full upload.
Four workflows where sensitive data leaves a system without protection applied, and where Pipeline closes the gap inline.
Case files must reach court, opposing counsel, or regulators with PII stripped. Paralegals spend days on manual redaction, and overlays can be stripped back off.
Documents route through the pipeline before production. Every file is scanned, all PII redacted, and a submission-ready set delivered with a full processing log. No manual review, no overlay risk.
Documents with PII or PHI are queued for chunking, embedding, and ingestion into a vector DB or RAG pipeline. The raw data has sensitive content that would enter the model.
Pipeline sits inline between source and embedding service. Files are tokenized before ingestion, so raw PII is never embedded or retrievable. Workflows get approved; analytical value is preserved.
Files generated in one internal system must reach a partner portal, a regulator, or a downstream app. Each transfer ships sensitive data with nothing applied at the moment it moves.
The application triggers Pipeline before a file reaches its endpoint — invisible to the user. The file that arrives has already had its sensitive content encrypted or tokenized. The workflow doesn't change.
Scheduled jobs push thousands of claims documents to payers and clearinghouses. Each batch carries PHI in plaintext, and the export job has no protection step built in.
A folder watch sits on the export staging directory. Every batch file is scanned and protected before it ships — PHI tokenized for the payer's systems, format preserved so downstream processing still runs.
File types supported, original format preserved
Protection modes include encrypt, redact, and tokenize
Data Solution of the Year for Legal, Data Breakthrough Awards
DLP tells you when something moved that shouldn't have. Pipeline protects the file before it moves — so even if it reaches the wrong destination, the sensitive content is inaccessible. DLP is detection; Pipeline is protection. DLP is actually a good reason to add Pipeline: you've already decided this data matters enough to monitor. Now protect it so monitoring isn't the last line of defense.
Full-file encryption makes the document unusable until decryption — anyone with the key sees everything. Pipeline applies selective encryption at the entity level. Only the SSN is protected; the rest of the document stays readable. Authorized users access what they're allowed to see, and AI systems can process the non-sensitive context. That's the difference between a hammer and a scalpel.
The analysis engine returns a confidence score with every detection. Policies can act only above a confidence threshold, so a low-confidence match doesn't trigger protection automatically. Entity definitions are fully customizable — if a default pattern is too aggressive for your content, it's tunable, and every detection decision is visible in the monitoring view.
Pipeline exposes a standard REST API — if your application can make an HTTP call, it can trigger Pipeline. Deployment is Docker or Kubernetes, with configuration files, Helm charts, and Docker Compose files provided. For the simplest use case — a folder watch with drag-and-drop — it's usable the same day.
Different jobs. Shield scans your data estate on a scheduled or continuous basis, handling what already exists at scale and keeping it encrypted and protected in place. Pipeline handles files at the exact moment they move. A common pairing: Shield holds the authoritative protected version of a document; Pipeline creates a clean copy for external distribution, redacting the sensitive fields so the file can go to a court, a regulator, or a partner while the original stays intact and access-controlled under Shield. Files created between scan runs, documents flowing through an application pipeline, or anything that needs protection at the instant of transfer are Pipeline territory. They cover different moments in the data lifecycle, not the same one.
Pipeline deploys in your environment in hours, with keys you control. No agents. No re-architecture. Protection at the moment of movement, before any file arrives.
One platform, no gaps with Discover, Shield, Pipeline, Exchange, AI Guard.