Protecto Vault masks sensitive data in large datasets using async APIs, built-in queuing and batch processing. Your AI pipeline gets tokens it can still use.
SOC 2 Type II · ISO 27001 · HIPAA, BAA available
Long texts processed per day
Lower cost than its in-house estimate
Claims documents in scope at a healthcare data platform
A SaaS team wants to train an AI agent on customer conversations. Masking works in the pilot. Then the full feed arrives, long texts pile up behind the masking step, and training starts late.
The model is waiting on the masking job.
Long texts take time to scan. Without a queue or batches, work piles up at peak and everything downstream waits.
Volumes rise and fall. Size for the peak and you pay for idle capacity. Size lower and you fall behind.
A miss rate that looks small in a pilot adds up to many exposed values at full volume.
Submit a job and get a tracking ID back right away. Your pipeline keeps moving, checks the status on a schedule and reads the result when the job finishes.
Failed jobs report a clear status and error message, so your code decides what to do next.
Submit the job
PUT /mask/async → tracking_id
Poll every 30 to 60 seconds
PUT /async-status → PENDING → SUCCESS
Read and store the result
result holds the tokens. Job data is purged later.
Async jobs need a paid plan. Book a demo to set up a proof of concept or trial.
These are the parts you would otherwise build and run around a masking library.
Use sync calls for live traffic and async jobs for large datasets. Both take the same request structure.
Large jobs wait in a queue and run in batches. This reduced peak-load delays for the SaaS company.
Resources follow demand, so you stop sizing for the peak by hand. The SaaS company replaced manual scaling this way.
Protecto retries internal processing errors up to three times. Masking is deterministic, so resubmitting the same data is safe.
50+ entity types across PII, PHI and PCI, plus custom entities, officially tested in seven languages.
Run Protecto on your servers or in your private cloud, or use SaaS. Multitenancy is included either way.
Speed only helps if the output is still useful. Redaction removes the words a model needs. Protecto swaps each value for a token that stays the same across every record, so the model can still tell who is who.
Ask the modelDid one customer contact us twice?
Basic redactionCan’t tell
[REDACTED] filed a claim.
[REDACTED] asked for a callback.
Protecto tokensYes, same person
[PERSON_7a1] filed a claim.
[PERSON_7a1] asked for a callback.
“Protecto enabled us to deliver data security, privacy, and compliance for our AI features at scale across 5,000+ enterprise clients, strengthening customer trust in our AI automation solutions.”
A masking library is one piece of the job. At volume, your team also owns everything around it.
| What the job needs | Build in-house | Protecto |
|---|---|---|
| Queuing and batches | You build and run them. | Built in. |
| Scaling | You write and maintain the scaling logic. | Auto-scales with volume. |
| Failed work | You design retries and error handling. | Internal retries, plus a status and error for each job. |
| Detection | You pick, tune and retrain models. | 50+ entity types plus custom entities. |
| Free text | Placeholders can strip the context AI needs. | Consistent tokens that keep context. |
| Operating cost | Your team carries build and upkeep. | 90% below the in-house estimate at one SaaS company. |
Independent benchmark: UT Dallas and DataXpert measured a higher F1 for Protecto than for AWS Comprehend and Microsoft Presidio.
Replacing sensitive values with tokens across large datasets, such as millions of long texts a day, using queues and batch jobs so pipelines keep moving. In Protecto, this is how Protecto Vault handles large workloads.
One SaaS company processes 13M+ long texts a day. A healthcare data platform has about 750M claims documents in scope. Bring your own volume and we will walk through how Protecto would run it.
Use async for large datasets and batch jobs. Use sync for live traffic such as model input and output.
Submit a job and receive a tracking ID. Poll the status endpoint every 30 to 60 seconds. When the status is SUCCESS, read the result and store it on your side, since job data is purged later.
Async APIs aren’t on the self-serve trial and need a paid subscription. To test them on your data, book a demo and we will set up a proof of concept or trial.
The status changes to FAILED with an error message, and your code decides whether to resubmit. Protecto retries internal processing errors up to three times, and resubmitting the same data is safe because masking is deterministic.
It can when placeholders carry no meaning. Consistent tokens keep those relationships.
On your servers or in your private cloud, or as SaaS. When Protecto runs in your environment, token mappings and keys stay on your servers.
A health analytics company ran its proof of concept in under a week. Book a demo and we will scope a proof of concept or trial for your volume.
Usually, yes. Tokenization is pseudonymization, not anonymization: while a mapping exists, a token can be linked back to a person. Your counsel decides what that means for each AI path.
Bring a data sample, with test data if you prefer, and your daily volume. We will set up a proof of concept or trial for your pipeline.
A third-party study compares how accurately Protecto, Microsoft Presidio, and AWS Comprehend find personal and health data in unstructured text.
By submitting you agree to Protecto’s privacy policy.
