Use code for 1,000 extra pages on your first month of Starter.+1,000 pages on Starter
See Starter →See →HyperAPI ranks first for redacting documents by API: masked pages, a [REDACTED] text block and counts per type from one call. Six tools compared.
Short answer: if you redact documents by API, HyperAPI by Hyperbots is our pick. You send a scanned or born-digital PDF or a page image (PNG, JPEG, WebP or TIFF), and one call returns the masked page images, the page text with each value replaced by [REDACTED], and a count of what was masked by type, at $50 per 1,000 pages with a free tier. Below, six tools are ranked on what goes in, what comes back and what it costs.
Published by HyperAPI. We rank ourselves first on the criteria below; check them against your own documents. Last updated: 18 November 2026. Vendor prices and docs read on 18 November 2026.
Try it on your own file. Sign up free for 30 pages a month with no card, and redact from the dashboard or the API. The request and response are in the redact API reference.
One limit up front, because it decides many projects: HyperAPI does not detect Social Security numbers, dates of birth, or card and account numbers by default. If those are what you must remove, read the sections on Azure, Google and AWS before you choose.
| # | Tool | You send | You get back | List price | Best for |
|---|---|---|---|---|---|
| 1 | HyperAPI (redact) | A PDF (scanned or born-digital) or a page image (PNG, JPEG, WebP or TIFF) | Masked page images, a [REDACTED] text block and counts by type, in one call | $50 per 1,000 pages | Masked pages from scans and page images by API, with black boxes or synthetic text |
| 2 | Azure AI Language (document PII) | PDF, Word or TXT in Azure Blob Storage | A redacted copy of the file and a JSON of entities, written to your container by an async job | $10 per 1,000 pages | Regulated identifiers in PDF and Word files, at the lowest page price here |
| 3 | Google Sensitive Data Protection | An image (JPEG, BMP, PNG) or text | The image with opaque boxes, or findings with locations | $3 per GiB inspected; first 1 GiB a month free | The widest catalogue of country-specific identifiers |
| 4 | Presidio (open source) | Text or images, on your own machines | Anonymised text, or a redacted image | Free (MIT licence); you run it | Self-hosting, with full control of the recognisers |
| 5 | AWS Comprehend | UTF-8 text | Entity types with character offsets, or a masked copy of the text | $1 per million characters | Text you already have, inside AWS |
| 6 | Adobe Acrobat Pro | A PDF, by hand | A redacted, optionally sanitised PDF | Per-user subscription | One-off manual redaction |
This list is for developers who redact documents (contracts, claim files, intake forms, statements) through an API, not for teams masking strings in a database. So the criteria are about documents:
Every vendor fact comes from that vendor's public pricing page, API reference or docs, read on 18 November 2026. We did not run any paid vendor API, and we did not measure detection accuracy for any tool, including ours. How many names or numbers a tool misses depends on your documents, so test each shortlisted tool on a sample of yours.
| Criterion | HyperAPI | Azure AI Language | Google SDP | Presidio | AWS Comprehend | Acrobat Pro |
|---|---|---|---|---|---|---|
| Scanned PDFs | Yes | Yes | Not directly; Google publishes a sample deployment that splits a PDF into page images | As images, after you render the pages | No (text only; OCR first with Textract) | Yes, by hand |
| Photos and page images | Page images (PNG, JPEG, WebP, TIFF) | No (“digital images with embedded text aren't supported”) | Yes (JPEG, BMP, PNG; not TIFF for redaction) | Yes (image redactor) | No | After converting to PDF in Acrobat, by hand |
| Office files | Not listed on the redact reference; convert to PDF | Word (.docx) and .txt | Inspected, but redaction is not listed for them | No | No | After converting to PDF in Acrobat, by hand |
| Returns redacted pages | Yes: masked page images, plus a [REDACTED] text block and counts by type | Yes: a redacted file and entity JSON in your Blob container | Yes for images | Yes for images; text for text | No: offsets, or masked text from a batch job | Yes |
| Masking modes | Black box, or synthetic text (deidentify) | Character mask (default), entity-type label, or none | Opaque box; box colour can be set | Replace, redact, mask, hash, encrypt, or your own operator | Character mask or entity-type label | Box, with optional overlay text and codes |
| Built-in entity types | 7: person names, company names, emails, addresses, websites, phone numbers, credentials | Includes US SSN, card and bank account numbers; date of birth in preview | 200+ infoType names, including US SSN, card numbers, date of birth, IBAN | Includes US SSN, card numbers, IBAN, dates and many national IDs | 36, including SSN, card and bank account numbers, dates | Search by words, phrases or patterns |
| Custom types | Yes, by name and description in pii_config | Choose categories with piiCategories | Yes (custom infoTypes) | Yes (regex, deny-lists, code) | Choose from the built-in types | Your own search terms |
| Price per 1,000 pages | $50 | $10 | Billed per GiB, not per page | Free; you pay for the servers | About $4 with Textract OCR (see the AWS section) | Not priced per page |
| Free tier | 30 pages a month, no card | 5,000 text records a month (text PII) | 1 GiB inspected a month | Free software | 50,000 units (5M characters) a month for 12 months | Free trial |
Azure wins two of these rows outright: price per page and regulated identifiers. HyperAPI ranks first because it is the only managed API here that takes both scanned PDFs and page images and returns masked pages, a redacted text block and per-type counts from a single call, with a synthetic-text mode and a free tier that needs no card.
HyperAPI is a document API with five operations on one key: parse, extract, classify, split and redact. Redact costs $0.05 a page in either mode. You upload the file, pass the returned document_key to POST /v1/redact, and the response carries three things:
output=urls as signed links for up to 60 pages. The links expire; store the request_id and fetch the result again for fresh ones.updated_text_block. The document's text with each masked value replaced by [REDACTED], ready to index or pass on.summary. A count per entity type, for example {"PERSON_NAME": 2, "EMAIL": 1}, which you can log as an audit record.mode=redact (the default) draws black boxes; mode=deidentify puts synthetic values in their place, so a page still reads naturally for review or test data. include_logos=true masks detected logos as well.
The Python SDK installs from GitHub and does the upload for you:
# pip install "hyperapi-sdk @ git+https://github.com/hyprbots/hyperapi-sdk.git"
from hyperapi import HyperAPIClient
client = HyperAPIClient() # reads HYPERAPI_KEY from the environment
resp = client.redact("intake-form.pdf", mode="redact", output="urls")
result = resp["result"]
print(result["summary"]) # masked values, counted by type
print(result["updated_text_block"]) # the text, with [REDACTED] in place of each value
client.download_pages(resp, "redacted") # one masked image per page
client.close()API=https://apis.hyperbots.com
FILE=intake-form.pdf
# 1. Ask for an upload URL
UPLOAD=$(curl -sS -X POST "$API/v1/documents/upload" \
-H "X-API-Key: $HYPERAPI_KEY" \
-H "Content-Type: application/json" \
-d "{\"filename\":\"$FILE\",\"content_type\":\"application/pdf\",\"content_length\":$(wc -c < $FILE | tr -d ' ')}")
DOC_KEY=$(echo "$UPLOAD" | jq -r .document_key)
# 2. PUT the file there (no API key on this request)
curl -sS -X PUT "$(echo "$UPLOAD" | jq -r .upload_url)" \
-H "Content-Type: application/pdf" \
-H "x-amz-server-side-encryption: AES256" \
--data-binary @"$FILE"
# 3. Redact, adding one type of your own to the built-in set
curl -sS -X POST "$API/v1/redact?mode=redact&output=urls" \
-H "X-API-Key: $HYPERAPI_KEY" \
-F "document_key=$DOC_KEY" \
-F 'pii_config={"mode":"extend","types":[{"name":"MATTER_NUMBER","description":"Internal case reference such as M-2024-0193","has_synthetic_replacement":true}]}'The response, as the API reference documents it (trimmed):
{
"status": "success",
"request_id": "req_01jb2...",
"task": "redact",
"result": {
"output": "urls",
"pages": [
{ "page_number": 1, "image_url": "https://files.hyperbots.com/...redacted/0.png?sig=..." }
],
"updated_text_block": "Invoice\n\nBill To: [REDACTED]\nContact: [REDACTED]\n...",
"summary": { "PERSON_NAME": 2, "EMAIL": 1, "ADDRESS": 1 }
}
}The built-in set is seven types: PERSON_NAME, COMPANY_NAME, EMAIL, ADDRESS, WEBSITE, PHONE and CREDENTIALS (passwords, API keys, tokens, secrets). It does not detect Social Security numbers, dates of birth, or card and account numbers by default.
pii_config lets you add types of your own, each with a name, a description and whether deidentify should substitute synthetic text for it. mode: "extend" adds them to the built-in seven; mode: "replace" uses only yours. We make no claim that a custom type catches SSNs or card numbers. If those must go, test that on your own documents, or use one of the tools below for them.
[REDACTED] text for search indexes and AI tools, and counts for your audit log.request_id, then deleted, and are not used for training. Sending the same document with the same settings within 24 hours returns the stored result at no charge; force_refresh=true re-runs it and is billed. HyperAPI is SOC 2 Type II.Where it is not the right pick: you need SSNs, dates of birth or card numbers caught out of the box; your files are mostly Word documents and you want Word back; a single document runs past 60 pages (split it first); or $50 per 1,000 pages is too much for your volume. The Python SDK installs from GitHub, not PyPI; any language works over REST.
For how this fits legal, insurance and accounting work, see HyperAPI by industry.
Azure's PII feature has three forms: text, conversations and “native documents”. The document form takes .pdf (including scanned PDFs), .docx and .txt from a source container in Azure Blob Storage and writes a redacted file plus an entity JSON to a target container. You start an async job and poll it. A request holds up to 40 documents and 10 MB in total, and the documented request uses a preview API version (2024-11-15-preview).
It wins on price: document PII redaction lists at $10 per 1,000 pages, the same in every region in Azure's retail price list. Its entity list includes US Social Security numbers, card numbers and bank account numbers, with date of birth in preview, and you pick categories with piiCategories. Masking is a character mask (the default), the entity type, or none. The limits: no photos or page images (“digital images with embedded text aren't supported”), tables in scanned documents are not supported, there is no synthetic-text mode on the document page, and you need a Language resource and two storage containers before the first call.
Best for: teams already on Azure who must remove SSNs or card numbers from PDF and Word files at volume.
Google's service (its API is still the DLP API) documents more than 200 built-in infoType names, from US_SOCIAL_SECURITY_NUMBER, CREDIT_CARD_NUMBER and DATE_OF_BIRTH to national IDs for dozens of countries, plus custom infoTypes. image.redact takes a base64 image, finds text by OCR and returns the same image with opaque rectangles over each finding. Redaction is listed for every supported image type except TIFF.
PDFs are the gap. PDF, Word, Excel and PowerPoint files can be inspected, but redaction is not listed for them. Google's answer is a sample deployment on GitHub (dlp-pdf-redaction) that splits a PDF into page images, redacts each one and reassembles the file, using several Google Cloud services you deploy yourself.
Pricing is by bytes, not pages: image redaction is billed as content inspection, free for the first 1 GiB a month, then $3 per GiB (and $2 per GiB above 1 TiB), with a 1 KB minimum per request. What that comes to per page depends on your image sizes.
Best for: country-specific and regulated identifiers in images, inside Google Cloud.
Presidio is the open-source PII toolkit that started at Microsoft (its GitHub address now redirects to the data-privacy-stack organisation). It is MIT-licensed. The analyzer finds entities in text using named-entity recognition, regular expressions, checksums and context; the anonymizer replaces, redacts, masks, hashes or encrypts them, or runs your own operator, such as a fake-data generator for synthetic values. The image redactor reads text from images with Tesseract OCR and draws boxes over what the analyzer finds; there is also a DICOM redactor for medical images.
Its entity list includes US SSNs, card numbers, IBANs, dates and many national IDs, and you can add recognisers. The cost is the work: you run and scale it, turn PDFs into images, and tune recognisers for your documents. Presidio's own docs say there is “no guarantee that Presidio will find all sensitive information”. That is true of every tool here.
Best for: data that cannot leave your network, and teams that want to own every detection rule.
Comprehend's DetectPiiEntities takes UTF-8 text (up to 100 KB per synchronous call) and returns each entity's type, a score and character offsets. An async batch job returns a copy of the text with each entity masked by characters or replaced by its type. It knows 36 entity types, including SSN, card and bank account numbers, passport numbers and dates.
Detect PII costs $0.0001 per 100 characters, with a 300-character minimum per request: $1 per million characters. AWS's own pricing example assumes 2,500 characters a page. At that density, Detect PII is $2.50 per 1,000 pages, and if the pages are scans, Textract text detection adds $1.50 per 1,000, for about $4 in total.
That buys offsets and masked text, not redacted pages: to mask the page images you match offsets back to Textract's word positions and draw the boxes yourself. The free tier is 50,000 units (5M characters) a month for 12 months from your first request.
Best for: text you already hold (chat logs, tickets, extracted fields) inside AWS, when you need SSNs and card numbers.
Acrobat Pro (redaction is a Pro feature, not Standard) lets you mark text and images, or search for words, phrases or patterns, then apply the redactions and choose whether to “sanitize and remove hidden information”. Adobe lists metadata, attachments, comments, form field data and hidden layers among the data Acrobat can redact. Its Action Wizard can run redaction across multiple PDFs at once, but Acrobat is still a per-user desktop subscription, not an API.
Best for: a lawyer, paralegal or analyst redacting a handful of PDFs by hand.
| If you need | Pick |
|---|---|
| Masked pages from scans, page images and PDFs by API, with names, contacts, addresses and credentials removed | HyperAPI |
| SSNs, card or account numbers removed from PDF and Word at the lowest page price | Azure AI Language (document PII) |
| Country-specific IDs in images, on Google Cloud | Google Sensitive Data Protection |
| Nothing leaves your network; no licence fee | Presidio |
| PII offsets in text you already have, on AWS | AWS Comprehend |
| A few PDFs, redacted by hand | Adobe Acrobat Pro |
A black rectangle drawn over a PDF does not always remove the text under it. We tested common ways of redacting PDFs and tried to read the text back in Can a redacted PDF be unredacted? HyperAPI's redact returns page images and a separate text block with [REDACTED] in place of each value, so there is no text layer under its boxes to copy out. Whatever tool you use, check the output file before you send it.
For redacting documents by API, HyperAPI: send a scanned or born-digital PDF or a page image (PNG, JPEG, WebP or TIFF) and one call returns masked page images, a [REDACTED] text block and counts by type, at $50 per 1,000 pages with 30 free pages a month. For SSNs and card numbers in PDF and Word files, Azure AI Language ($10 per 1,000 pages). For free self-hosting, Presidio.
Azure AI Language, Google Sensitive Data Protection, AWS Comprehend and Presidio all have a built-in US SSN type. HyperAPI does not detect SSNs, dates of birth, or card and account numbers by default; its built-in types are names, companies, emails, addresses, websites, phone numbers and credentials.
HyperAPI's free plan includes redact: 30 pages a month, no card. Google gives 1 GiB of inspection a month, AWS Comprehend 5M characters a month for 12 months, and Azure 5,000 text records a month for text PII. Presidio is free open-source software you run yourself.
At list price, HyperAPI is $0.05 a page and Azure document PII $0.01 a page. Google bills $3 per GiB of images inspected, and AWS Comprehend $1 per million characters of text, so their per-page cost depends on your files.
Redaction hides a value, usually with a black box. De-identification replaces it with something that is not the original, such as a synthetic name, so the document still reads naturally. HyperAPI does either: mode=redact or mode=deidentify.
18 November 2026: first published. Vendor pages read on 18 November 2026.
Start free with 30 pages a month and no card, or read the redact API reference.