Back to workspace
DAIDA / PRODUCT GUIDE

DaidaRedact Help

Upload your project, approve the estimate, and let Azure finish the work. Return whenever you need to check progress.

Version 2.0.6Unattended Azure editionOctober 8, 2026
After uploading, the work runs in Azure.

Keep the tab open while files are still uploading from your device. Once uploading is complete, preparation can continue without your browser. Redaction waits for your explicit approval of the page count, estimate, entities, and style. After starting the project, you can sign out or turn off your computer.

1. Start a project

  1. Create a project and add PDFs. Drop files or browse your device. Files upload directly to the configured private Azure Blob source container. A filename does not need a .pdf extension; its contents must be a readable PDF. This edition submits documents as English.
  2. Finish uploading, then prepare. Background workers validate the files, count pages, and split large PDFs into suitable processing parts. Preparation does not submit a redaction request. Unreadable files are listed for attention.
  3. Choose the information to redact. No PII entities are selected automatically. Choose the entities and redaction style you need; for example, Person and PhoneNumber with Black out.
  4. Review and approve the estimate. The estimate uses counted original pages and the configured rate. At $0.01 per page image, enter $10 per 1,000 pages; 10 pages estimates to $0.10. Splitting does not multiply the original-page count. Changes to settings or reviewed files require a fresh approval.
  5. Start once, then check as needed. The app sends groups of up to 20 PDFs or processing parts, within 10 MB combined per request, and handles status checks, reassembly, and reporting in Azure. Multiple requests can process concurrently. You can upload a larger project and let it continue overnight; the app manages the smaller groups for you.

The estimate is a planning amount, not an Azure invoice. Storage, hosting, transfer, retries, and taxes can add cost. Your administrator configures Azure once at deployment; individual users do not enter a Language key or container credentials.

Files already uploaded to Azure

An administrator can select a customer folder in the configured source container. Complete that folder upload first and keep its contents unchanged during preparation. Use a dedicated customer folder, not the container root or the app’s internal folders. This capability is restricted to administrators because it can access shared source files.

Amazon S3 import is not part of this edition.

2. Sessions, interruptions & recovery

Will signing out stop combining?

No. Background workers download the redacted parts, combine them in page order, validate the result, and save the completed PDF. Your login session is separate from that work.

Signing out stops your dashboard access. It does not pause an already-started project. Use the project’s Pause control when you want processing to pause.

What still needs an open upload tab?

Bytes that have not yet left your device cannot upload after the tab closes or your computer sleeps. Wait for upload completion. If transfer is interrupted, sign in again and resolve incomplete uploads before preparing the project.

Already saved files and project progress remain in Azure. You do not need to keep a browser open for page counting, redaction, or reassembly.

Sessions expire after 30 minutes without authenticated requests or eight hours total. Sign in again to see current progress. Account disable, password reset, and sign-out revoke access; they do not cancel a previously approved background job.

Workers save checkpoints and coordinate retries. Transient failures can recover automatically. If Azure may have accepted a submission but its reference was not saved, the app stops automatic resubmission for that work to avoid duplicate processing. Ask an administrator to reconcile the Azure outcome before proceeding. Retry rejected submissions only when Azure definitively rejected them.

A pause takes effect at a safe checkpoint. Azure requests already accepted can finish. The optional recovery sweep runs every five minutes to find persisted work whose queue notification was missed.

3. How records move

Each original stays associated with its processing parts and final output. Large projects are divided into internal groups automatically.

  1. Your device → private Blob sourceSave the original

    Upload the entire project. An administrator can also select a completed source-container folder upload.

  2. Background workerCount and prepare

    Check PDF content, count original pages, and split files on page boundaries where necessary.

  3. Your project dashboardApprove pages, cost and settings

    Review exclusions and choose PII entities and redaction style. Paid redaction starts only after approval.

  4. Work queue → Azure LanguageSubmit automatically

    Workers send groups that fit the service limits. New submissions can overlap accepted Azure jobs; throttled work waits and retries.

  5. Azure Language → Blob outputDetect and redact

    The service writes redacted documents and detection JSON. The worker saves status without requiring your session.

  6. Background worker → Blob outputReassemble and verify

    Combine successful parts in their original order and verify the page count. Incomplete results are withheld.

  7. Blob output → your teamDownload and review

    Use the project report, PSV or JSON detections, delivery manifest, and completed PDFs. Bulk-download through Storage Explorer.

  8. DocumentSync → project cleanupVerify delivery, then purge Azure

    Confirm the imported files and pages in DocumentSync. An administrator then removes that project’s Azure records and files.

4. Architecture & technology stack

Your browser

Upload PDFs · choose entities and style
Approve the estimate · check results

DaidaRedact portal

Sign-in and user administration
Secure uploads and project controls

Starting a project saves durable work in Azure
BACKGROUND PROCESSING · CONTINUES WITHOUT YOUR BROWSER
Work queue + saved state

Work waits safely between steps
A recovery sweep finds missed work

Background workers

Count · split · submit · check
Reassemble · validate · report

Workers coordinate the document service and storage
Blob source

Original PDFs and smaller processing parts

Azure AI Language

Document-based PII detection and redaction

Blob output

Redacted files, final PDFs and project reports

Download → verify in DocumentSync → manually purge project files
The portal and workers are separate. Closing a completed upload tab or signing out does not stop preparation, redaction, reassembly, or reporting. Accounts and project checkpoints persist outside the running containers.
Technology used by the unattended Azure edition
LayerWhat runs there
PortalReact and TypeScript with Next.js on Node.js 24, hosted in Azure Container Apps Consumption. HTTPS requests wake the portal when it has scaled to zero.
User accountsApp-managed email/password accounts, roles, password hashes, sessions, and shared sign-in controls in Azure Table Storage. They persist independently of the portal.
Work coordinationAzure Queue Storage, saved Blob project records and leases, and a scheduled recovery sweep. The browser displays status; it does not drive processing.
PDF preparationContainer Apps Jobs use Node.js and pdf-lib to count, split, and combine PDFs. Workers run independently of web requests and login sessions.
RedactionAzure AI Language Document-based PII through its asynchronous API. The app does not make a separate Azure Document Intelligence call.
Storage accessManaged identities access private Azure storage. Short-lived, file-scoped links allow uploads/downloads. The Language resource uses its own system identity for input/output access.

No dedicated VM is required. Portal and event-worker compute can scale to zero. Retained storage, the registry, logs, domain, and the periodic recovery check can still incur costs. A cold portal visit may take longer to load.

5. Reports, confidence & delivery

The worker automatically creates one comprehensive project report after processing. Choose PSV for a structured pipe-delimited export or JSON for detailed detection data. The delivery manifest maps original filenames to completed output locations. Individual PDF detection reports do not have to be opened one at a time.

What confidence means

The review threshold flags detections below the chosen confidence level. A missing confidence value stays unknown. It does not become zero, and it does not prove that a file contains no PII.

This review threshold changes report flags; it does not undo redactions or change Azure’s detection behavior. Review the outputs before sharing them.

Large and failed records

Split-PDF report rows retain their original-file and page mapping, so report rows can outnumber uploaded PDFs. Failed files and missing or unrecognized detection data are identified.

New submissions request the model selector latest. A recorded requested selector is not proof of the exact model Azure used. Unknown model records stay unknown.

Detection reports can contain original names, phone numbers, and other PII.

Protect PSV/JSON reports and raw Azure detection data as carefully as source PDFs. A redacted PDF and its detection report have different disclosure risks.

Download a large project

Use Microsoft Azure Storage Explorer with an Azure account that has access to the output container. Locate the project’s output folder and download its PDFs, reports, and delivery manifest. This avoids clicking every PDF in the browser. Your DaidaRedact app password is separate from Azure Storage Explorer sign-in.

Files with matching names are kept in separate file-ID folders. Use the delivery manifest to map them back to the originals. Import the downloaded records into DocumentSync and verify the expected files, page counts, and readable results before cleanup. This edition does not automatically import into DocumentSync.

6. Storage & retention

LocationWhat is kept
Source containerUploaded originals, prepared copies and split parts, internal batch/job records, and child-report preparation data.
Output containerAzure’s raw redacted documents and detection JSON, reassembled PDFs, final project reports, and delivery manifests.
Control containerProject ownership, selected files, saved progress, approved settings, and coordination records.
Account tableUser accounts, password hashes, roles, sessions, and sign-in controls. Keep these for future projects.

Azure files are not purged automatically when a job finishes. After DocumentSync delivery has been verified, an administrator must remove that project’s originals, split inputs, raw and final outputs, reports, internal batch records, and control records. Keep the project record long enough to identify every related storage path.

Do not delete whole shared containers, another project’s files, the account table, or the resource group as ordinary project cleanup. If soft delete, versioning, snapshots, or retention policies are enabled later, retained copies can remain billable after a normal delete. The supplied deployment disables Blob soft delete and versioning on its newly created storage account.

7. Troubleshooting

The portal opens slowly after time away

The portal may be starting from zero running instances. Allow it to finish starting. This does not affect a background project.

The project is waiting for approval

Page preparation has finished, but redaction has not started. Choose at least one PII entity, review the style and estimate, then approve and start.

Nothing progresses while I am away

Check whether the project is paused, waiting for approval, or needs attention. An administrator can inspect worker executions, the queue, storage permissions, and the recovery job. Signing back in displays progress; it is not required to drive the worker.

Workers are active but completed documents are not increasing

The last worker check confirms activity, while the last document outcome change records newly completed or failed documents. These are different signals. Failed-document counts and long-running-job alerts appear while other work continues. Choose View Azure processing details to inspect saved job IDs and status timestamps without retrying redaction.

Jobs pending longer than the configured alert threshold (two hours by default) are flagged for administrator review. This is an alert, not an Azure deadline. The app continues polling the original Azure job reference and does not automatically submit a duplicate. Version 2.0.3 also remembers which internal batch should be visited next, so later batches receive processing time across worker runs.

Azure rejected a request

Choose View Azure processing details, then Copy Azure processing details. Use Check remaining groups if shown. This reads saved errors across document pages without retrying redaction. Copy any Azure code, field and request ID provided. An administrator should check the Language endpoint/key, supported policy/API, readable PDF input, and Language system-identity access to source/output containers. Do not include keys, access links, or document contents when sharing diagnostics.

An accepted job keeps running but its status check fails

A permanent HTTP 400, 401, 403, 404, 410 or 422 status-check error stops repeated polling and puts the project in Needs attention. Choose View Azure processing details to see the accepted Azure job ID, failed check, last successful poll, safe error codes and request ID. A 400 response alone does not prove that the result expired. Do not use rejected-submission retry for an already-accepted job.

Check files Azure has already saved

For an approved project that is paused or needs attention, choose Check saved outputs and acknowledge the review requirement. The background worker checks the saved Azure job status first. If it still cannot retrieve status with HTTP 400, 404, 410 or 422, it inspects that job’s exact document output paths. It never sends a new redaction request. HTTP 401/403 requires correcting access first.

Recovery requires a readable PDF with the expected page count and an accompanying recognized detection report. The output must stay unchanged during checking and delivery. Recovery recognizes shared output folders within the accepted Azure job and allows recovered PDFs up to 128 MB. Azure outputs can be larger than their inputs; this does not change the submission size limit. A unique job and document match with a paired detection report is required. Detection reports can identify entities using category or type; version 2.0.6 supports both formats. Reports with incomplete entity records remain withheld. A match by filename alone is insufficient. Missing, ambiguous, changed, unreadable or unsupported outputs remain withheld.

Recovered files carry a review warning: file integrity checks do not establish Azure’s terminal status or certify that every sensitive item was redacted. The pass pauses when finished. If any parts remain unresolved, keep the project paused, review their errors and correct the issue before choosing Check saved outputs again. When no parts remain unresolved, review any recovered PDFs before using Resume to continue normal processing. If a pass was interrupted, Resume continues its saved checkpoint. The completed count includes delivered recovered files; their review warnings remain visible. Checks and downloads can still incur storage, transfer and worker charges.

Azure says too many records or Max 20 records

Version 2.0.2 limits each Azure request to 20 PDFs or processing parts. After your administrator updates the portal and all workers, open the existing project and choose Retry rejected groups. The app divides eligible saved groups into smaller requests; you do not need to upload the files again. Previously accepted work keeps its existing job references. An uncertain submission still requires an administrator to check Azure before retrying.

The project page reports HTTP 503

A failed dashboard request is separate from a rejected Language submission. Copy the displayed support reference. An administrator can match it to the web app logs and check identity, storage access or service availability. Refreshing status does not resubmit redaction.

A PDF cannot be prepared or combined

Use a readable, unencrypted PDF. A single page that still exceeds the service size limit needs its image size reduced before upload. Very large originals can exceed the configured worker memory limit. Failed or missing redacted parts prevent release of an incomplete combined PDF.

What happens to the original?

The uploaded original is preserved. Split and reassembled derivatives may not preserve document-level bookmarks, forms, attachments, tags, or digital signatures. Keep the original wherever those features matter.

Reference guides

Installation and operating checks are in DEPLOY-AZURE.md in the deployment package. Download help saves this guide and its diagrams as a self-contained HTML file; Print / save PDF uses your browser.