Practical document tool
Anonymize & Redact uses local text extraction, OCR and QR/barcode decoding to suggest regions for human review. Content marked Redact is removed from newly reconstructed pixels, and export is blocked when final verification detects recoverable information marked Redact.
Detailed product guidance
Anonymize & Redact uses local text extraction, OCR and QR/barcode decoding to suggest regions for human review. Content marked Redact is removed from newly reconstructed pixels, and export is blocked when final verification detects recoverable information marked Redact.
Read the full Anonymize & Redact guideAnonymize & Redact helps remove selected personal, contact, identity, financial and machine-readable information before sharing a document. Local detectors suggest regions. You decide what to Redact or Keep visible and add manual redactions for anything missed.
Anonymization here means removing the information you select. It is not a guarantee that a document can never be associated with a person: context, unusual events and information you leave visible can still identify someone.
Sign in with an eligible ALTABRIG account. Preserve the original. Use a synthetic example first and review every page before sharing an output. Automatic detection is not a substitute for your judgment.
| Plan | Shared actions | Effective input | Transcribe | Priority |
|---|---|---|---|---|
| Guest Free | Processing denied | — | — | — |
| Registered Free | 5 credits / rolling 24 hours | 10.48576 MB/action | Uses the shared action pool | standard |
| Pro | Unlimited standard actions | 10.48576 MB/action | Uses the shared action pool | priority |
| Pro Plus | Unlimited standard actions | 10.48576 MB/action | Uses the shared action pool | highest |
The initial engineering envelope is one file, at most 10 MiB and 5 pages, within your current central allowance. At 200 DPI, rasterization is limited to 12 million pixels per page and 20 million pixels across the document. Large page dimensions can exceed these limits even when the file is small. Each processing stage has a 120-second timeout and uses existing global queue capacity.
V1 accepts PDF, JPG/JPEG and PNG. Native-text, scanned and mixed PDFs use the same review workflow. Rotated and differently sized pages are processed according to their visible geometry. Password-protected PDFs are rejected.
DOCX, XLSX, PPTX, WebP and animated/multipage image formats are unsupported by this version. Convert only through a separately trusted workflow if you need to prepare a PDF first, then inspect the converted appearance carefully.
PDF output is a new image-only secure flattened PDF. Image inputs produce a new PNG. Neither output overwrites the uploaded original. There is no promise of editable text, a searchable text layer, preserved forms or lossless original encoding.
[REDACTED] under the top Redaction appearance controls. Acknowledge any digital-signature warning.The original remains a separate browser File during this session. Analysis and sanitization submit it through the owning user's queue. The server checks the original content digest against the signed workflow reference before exporting. Changing the source or the custom terms requires a new analysis.
Do not close or reload the workspace while reviewing: this version does not autosave the review plan. Choose another file resets the current workspace.
Detectors include email addresses, international phone candidates, configured Latvian-style identity candidates, contextual birth dates and document numbers, validated IBANs for configured countries, BIC/SWIFT candidates, validated payment-card candidates, payment/account references and your custom terms. OCR supports English, Latvian and Russian together.
IBAN checksum and payment-card Luhn checks filter invalid lookalikes. A matching pattern is still a review candidate, not a statement that a person, account or payment is authentic. Other countries or unusual formats can require manual redaction or custom terms.
Names and addresses are conservative contextual suggestions. Handwriting, blur, small print, dense layouts, low contrast and uncommon spellings can cause misses. No universal name/address recognition or perfect OCR accuracy is promised. The displayed confidence combines the detector's provenance with the source text or OCR confidence.
Every automatic finding starts as Needs decision. Redact marks it for removal; Keep visible leaves it visible unless another redaction overlaps it. The top status always identifies the remaining decisions and next action.
The Findings panel is a selection and status list. It shows category, page, confidence, source channel and review state without duplicating decision buttons. Select a row with a click or use the arrow keys, Home/End and Enter/Space. On narrow screens the top View menu opens Findings; selecting a row closes the drawer and returns focus to its top control.
Inspect the original page itself. Hidden or machine-readable material may require redacting the entire indicated region. Redact all remaining operates across every page, regardless of panel filters, preserving prior Keep visible decisions.
Under Redaction tools, choose Draw redaction, then drag over a region on the visible page. Alternatively, Add manual region creates a region that you can adjust with the keyboard.
A manual region is always set to Redact. Drag it to move it, use its corner to resize it, or edit x, y, width and height percentages in the selected-region controls. Remove manual redaction in the top contextual controls deletes it. Automatic finding geometry stays tied to the analysis; choose Keep visible for an automatic finding and add a manual region when you need a different selection.
The sanitization engine includes a small safety margin. It can expand a selection to include complete words or machine-readable regions that touch it. Leave sufficient separation from nearby content and inspect the resulting output.
Undo and Redo under Redaction tools restore recent changes to decisions and manual regions. Up to 50 recent review states are kept in this browser workspace. A new edit clears the redo path. Edits invalidate the previous download and require another verification.
Filter by class, review state, confidence below 75%, and current/all pages. Selecting a finding navigates to its page. Use Previous, Next, the page number input or optional thumbnails for deterministic page navigation. Filters do not remove findings or count as review decisions.
Zoom changes the review scale only. Fit page displays the complete page where space allows; Fit width prioritizes readable page width and may require vertical scrolling. Narrow layouts group Fit controls and secondary commands into menus before labels become cramped.
Full screen enlarges the document workbench. Page, findings, decisions and review history stay in the same session when you enter or leave it. The findings button opens the findings panel on narrow screens; close it to return to the document.
Advanced options is collapsed by default. Expand it with a click, Enter or Space. These values are optional: add names, case numbers, project names, account references or other exact text you want flagged. Enter up to 100 nonempty terms, each at most 160 characters. Matching is case-insensitive against available native/OCR text. It cannot find text that neither extraction nor OCR recognizes. Custom regular expressions are not supported. Use manual regions for unusual or non-text content.
The local reader checks QR codes and common 1D classes including Code128, EAN13 and Code39. Additional recognized symbologies can appear as candidates. Every decoded region remains reviewable; not every barcode contains sensitive data.
Redact the complete code region when it must be removed. Partial erasure is unsafe because error correction or remaining scan lines can still recover the payload. Manual regions touching a recognized code expand to cover that code. Verification decodes the final pixels again and blocks the output if a selected payload remains recoverable. Unsupported, damaged or extremely small codes may be missed; inspect the page and use manual redaction where needed.
The tool checks PDF signature objects where supported. A warning requires acknowledgement before exporting a derived copy: its bytes are changed and the original digital signature is not carried forward. The original signed bytes remain separate. The output does not retain stale signature fields that imply validity.
Scanned handwritten signatures are manual-only. Add a region covering the entire signature. Password-protected or encrypted PDFs fail closed; this tool does not unlock them.
Black is the default appearance. White creates blank-looking replacement pixels. [REDACTED] adds a static label where the region is large enough; small regions remain solid. All modes reconstruct pixels. A white region is just as intentional a removal as a black region, even though it is less visually conspicuous.
Preview anonymized document displays review overlays in the workspace. It is not a sanitized file and must not be treated as proof that the original is safe to share.
During export, the tool creates new page images, replaces selected regions in the pixels and embeds only those sanitized images in a new PDF. It does not copy the source PDF text/OCR layer, interactive objects, attachments or arbitrary metadata. Images are freshly encoded without source EXIF/GPS/device metadata. Page order and visible dimensions are retained within the verified rendering mode.
File creation alone is not success. The final PDF is independently rendered, checked with text extraction and qpdf, inspected for prohibited objects, and scanned with OCR and the local QR/barcode decoder. Final image metadata and pixels are checked as applicable.
If selected material remains recoverable, output is blocked. No unsafe artifact is offered with a warning. The check is conservative: repeated selected words or values elsewhere can block export even if one occurrence was redacted. Review and redact the remaining occurrences, then verify again. Verification covers the supported channels and selected regions; it does not prove that every sensitive fact in the document was detected.
Analysis and sanitization use the existing global durable queue. Review edits remain in the browser. One logical workflow shares the existing Credit Engine action reference so analysis and export do not create two customer actions. Retries and repeated downloads follow central idempotency/accounting rules; a new analysis is a new workflow. Queue waiting time can increase when native processing capacity is occupied.
document-sanitized.pdf for a PDF source: a new image-only PDF.image-sanitized.png for JPG/JPEG or PNG: a newly encoded image.The output filename does not reuse a potentially sensitive original filename. Ordinary PDF text search will not find text in the image-only output. Recipients can still run OCR on visible unredacted pixels.
Open the downloaded file in independent software. Inspect every page, orientation, label, redaction edge, signature and code. Check that necessary unredacted content remains readable. Share the sanitized download, never the original or a screenshot of the review workspace.
Document processing is local to the ALTABRIG service. No external OCR, AI, decoding service or CDN receives document bytes or barcode payloads. Findings and previews are available only through the owning user's workflow. Admin telemetry must not contain document previews, extracted values or decoded payloads.
Per-job native scratch is removed in finally cleanup after success, failure or cancellation. The shared queue temporarily stores request/result bodies using its existing restricted storage and terminal cleanup policy; it is not an immediate-erasure service. Its current default terminal retention is 24 hours, subject to the centrally configured policy and cleanup scheduling. Browser review memory lasts for the current session. No new persistent raw-finding table is used.
| Plan | Result | Metadata |
|---|---|---|
| Guest Free | Not retained | Not retained |
| Registered Free | Not retained | 1 day |
| Pro | Not retained | 30 days |
| Pro Plus | Not retained | 90 days |
For unsupported-type or MIME errors, choose an actual supported file rather than renaming its extension. For encrypted/damaged PDFs, obtain a valid unencrypted copy through your normal authorized process. For resource errors, use fewer or smaller pages. After a session expires or the source changes, analyze again. Cancel processing through the visible queue controls and allow cleanup to finish before retrying.
If verification blocks export, revisit all occurrences of the selected content and inspect complete machine-readable regions. Never bypass verification or substitute a rectangle drawn over the original PDF. If export is locked, follow its adjacent reason: resolve the remaining decisions, preview after the latest change, or acknowledge the digital-signature warning. Draft preview never unlocks export while decisions remain.
Synthetic contact sheet: Add EXAMPLE PERSON as a custom term, redact the synthetic email and phone, manually cover the signature, and verify the download.
Synthetic payment sheet: Review test IBAN/card candidates and redact a QR code containing a test reference. Confirm the code is absent from the verified output.
Can I undo after downloading? You can edit the browser review plan and create another derived output. Removed pixels cannot be restored from the sanitized artifact; keep the separate original securely.
Why did a code marked Keep visible remain? Keep visible intentionally leaves that candidate unselected unless another redaction overlaps it.
Does a successful check mean perfect anonymization? No. It confirms the supported export verification passed for the selected plan. You must assess undetected information and document context.