Document Processing · Control panel

PIPELINE

Reading documents

Choose how the scanner reads text and finds fields. Changes apply to new documents; work already running keeps its settings.

Settings

Approved patterns

Reuse tested layouts first, and keep AI suggestions as drafts for review.

Use validated patterns before metadata interpretation Automatically ask the selected AI to save drafts for unfamiliar layouts

Basic settings

Choose the text readers and whether fields come from approved patterns, text rules or your selected AI.

Service name

Main text readerUsed for images and scanned PDF pages.

Backup text readerUsed when primary extraction fails or is weak.

How fields are foundChoose rules, text only or your selected AI profile.

Active AI profileUsed for AI interpretation and pattern drafting when enabled.

Automatic extraction fallback

Use approved rules first. If a layout is unfamiliar, try a free local model, then configured paid models. Uncertain documents wait for review; no paid stage runs without its permission and key.

Enable template-based automatic fallback

Free local model

Cheaper paid model

Advanced paid model

Return additional source-backed details separately; the Hub ignores unmapped fields Save safe reusable learning drafts for administrator review

Advanced limits and quality

These limits control file size, reading quality and processing time. Keep the current values unless you need to change how documents are processed.

AI request timeout (seconds)

Maximum AI input charactersLarger documents keep OCR text and go to review.

Maximum AI output tokens

OCR quality thresholdA quality signal; it does not verify staff identity.

Maximum upload (MiB)

Maximum pages

Processing timeout (seconds)

PDF rendering resolution (DPI)

EXTENSIONS

Technology providers

Independent processing stages keep future features separate from your client integration.

MODULAR PIPELINE

+

Independent AI adapters

Configure multiple provider/model profiles below. OCR stays local; only extracted text goes to the AI profile you enable.

AI INTERPRETATION

AI providers & keys

Save profiles, add their separate keys, then select an active profile in processing configuration.

Private credentials

Choose an AI provider and model

A profile is a saved provider/model combination. You can add several and choose which one the scanner uses in Reading settings.

Edit saved profile

Add the provider API key

Select a saved AI profile and enter the key supplied by that provider. Keys are stored privately and are never shown back.

This sends a small synthetic example. A paid provider may charge for the test. No real documents are sent.

INTEGRATION

Connect an app

The client owns document storage, filing and manager approval.

EXTRACTION ENDPOINTPOST /api/extract

Raw document bytes · dedicated client API key · scanner source

Client credential

RETURNED METADATA

document_name
staff_name
document_type
expiry_date

Unknown fields are null. Text, page evidence and review reasons are returned alongside metadata.

Direct manager uploads inside the connected app bypass this service. Queued intake holds documents and results temporarily until the destination acknowledges receipt. The connected app handles staff matching, filing and approval.

Automatic folder/API intake uses POST /api/jobs with the care-home intake key and source ID. SFTP uses the care-home username and its configured source folder. The scanner delivers the original and JSON to the home's webhook.

What can the receiving app save?

Declare the document types, owner kinds and field formats supported by an app. These settings describe an existing integration; they do not create database columns in that app.

Receiving app capability

App IDApp name

Owner kinds supported by the app

Supported document types

Training topics use the training-certificate type; the course title distinguishes them. Layout variants are taught separately.

Writable destination fields

Use the receiving app's actual field names. Owner IDs are handled by its trusted integration, not by text extraction.

Choose what this client needs

Each registered connection has its own keys, sources and delivery route. Select its document types, map the fields it needs and mark fields required for filing. Missing information can be filled in by its receiving app.

Registered connectionReceiving appConnection nameEnvironment

Enable this requirements profile

Separate app installations need separate registered connection IDs and keys. Staff or applicant identity is validated by the receiving app; uncertain ownership stays unassigned.

Document type Accept this document type

Document belongs toDocument type ID in the receiving app

An empty destination ignores the field. Completion, issue, explicit expiry and next service due dates stay separate unless deliberately mapped.

LEARNING WORKSPACE

Document types & reusable patterns

Teach once, extract with rules next time. AI suggestions are saved as drafts. Samples stay in this open page and are cleared when you lock it.

Document template

Layout pattern

Fields for this document type

A document type defines which fields should be collected. For example, training uses course and completion date; identification uses document number and date of birth.

Type ID

Type name

Needed for filing marks a Hub requirement. Missing values are preserved as empty; the scanner never invents them. Changing fields pauses affected patterns until they are tested again.

Advanced: edit template fields

Template JSON · id, label, fields (id, label, kind, required)

Choose where fields go

Map captured fields to the receiving Document Hub. Training completion date can map to start date. These are delivery instructions; the receiving app must support the fields.

Applies to

Teach a layout

Load one sample, choose stable headings, then show which text belongs to each field. Save a draft before testing it. Personal names and dates must not become recognition headings.

Layout IDLayout name

Required recognition headings · one per lineUse at least two stable headings. Avoid names, dates and document numbers.

Load extracted result JSON or textTest an original document These images are pages of one document, in selection order

PageExtracted text · select a value to teach its fieldFor multi-page results, pages are preserved until you edit this text. Upload here tests the service; manager filing remains in the client.

Selected text belongs to

Select a value in the text, then choose its field.

Where is the field value?Stable label / headingStop at this label (optional)

Following lines to skip

Lines to read

Date order

Advanced layout JSON

Reusable layout JSONNo executable code is accepted.

Import a layout JSON as a draft

Test and approve a pattern

Check expected values, then collect two different matching examples and one unrelated example. Examples are used for this test only; the saved report contains pass/fail results.

Expected field values for current positive example

No examples collected

After activation, recheck matching held documents for

Only the selected care home is rechecked. Uncertain documents remain held; delivered documents are untouched.

Activate when these tests pass

Saved versions

Saved versions · activate a validated version to roll back

Automatic learning drafts

These proposed templates and rules are inactive. Review its capture fields and reusable headings before accepting it into the teaching library. It still needs two matching examples and one unrelated example before activation.

Unlock the panel to load drafts.

Preview and test results

This shows what the draft extracted and whether validation passed. A saved draft stays inactive until the required examples pass and you approve activation.

Load a sample and preview a pattern.

Advanced: full result JSON

Load a sample and preview a pattern.

Processing & delivery queue

Unknown layouts and unreliable extraction wait for administrator correction. Failed deliveries keep the document for retry. Originals and results are removed only after acknowledgement.

OCR administrator correction

Extracted OCR text

Folder connectors report in every 30 seconds. A missing report for five minutes shows offline. SFTP and API sources show their last upload instead.

Operations & source health

Unfamiliar documents

Alerts store document references and recognition status. Queued originals and OCR results wait in Care homes & delivery for administrator review. Transient API samples can be loaded in the teaching editor.

INTAKE & HANDOFF

Care homes, sources & destinations

Each care home has separate intake credentials, source locations and a receiving webhook. Originals are held until the destination confirms receipt.

Care home

Register a care home

The care-home ID identifies the sender. Its sources identify scanners or watched folders. Its destination tells this service where to send that home’s documents and JSON.

Care-home IDThis identifies the home in every upload. It cannot be changed after registration.

Care-home name

Enable this care home

Allowed SmartCare widget origins · one per lineExact HTTPS origin, without a path. Local HTTP origins are allowed for localhost, 127.0.0.1 or the explicitly configured development origin.

Keep a new home disabled until its sources, access keys and receiving app are ready.

Scanners and watched folders

Register each scanner or folder connector with its own source ID. Store the folder or SFTP location so you can see where a document came from.

Where results should go

Enter the receiving app's exact webhook URL. This service sends the original document and extracted JSON there after processing and any required correction.

Delivery method

Download mode holds completed originals until your app stores and acknowledges them. No webhook is required.

Receiving webhook URL (push mode only)

Use the dedicated receiving route, not the app's homepage. HTTPS is required outside local testing.

Who can send and receive

Save the home first. Keep its intake key with its folder connector. The receiving app needs the matching webhook signing key.

Start intake and delivery

Enable this only after care homes, access keys and receiving apps are ready. Unknown layouts still wait for administrator review.

Your processing workspace

Overview

How documents move

Reading documents

Approved patterns

Basic settings

Automatic extraction fallback

Technology providers

AI providers & keys

Choose an AI provider and model

Add the provider API key

Connect an app

What can the receiving app save?

Owner kinds supported by the app

Supported document types

Writable destination fields

Choose what this client needs

Document types & reusable patterns

Fields for this document type

Choose where fields go

Teach a layout

Test and approve a pattern

Saved versions

Automatic learning drafts

Preview and test results

Processing & delivery queue

OCR administrator correction

Details for this document type

Operations & source health

Unfamiliar documents

Care homes, sources & destinations

Register a care home

Scanners and watched folders

Where results should go

Who can send and receive

Start intake and delivery

Connect to your service

Overview

How documents move

Reading documents

Approved patterns

Basic settings

Automatic extraction fallback

Technology providers

AI providers & keys

Choose an AI provider and model

Add the provider API key

Connect an app

What can the receiving app save?

Owner kinds supported by the app

Supported document types

Writable destination fields

Choose what this client needs

Document types & reusable patterns

Fields for this document type

Choose where fields go

Teach a layout

Test and approve a pattern

Saved versions

Automatic learning drafts

Preview and test results

Processing & delivery queue

OCR administrator correction

Details for this document type

Operations & source health

Unfamiliar documents

Care homes, sources & destinations

Register a care home

Scanners and watched folders

Where results should go

Who can send and receive

Start intake and delivery