​

ADP Product Manual (SaaS)

​

Agentic Document Extraction

​

Key Concepts

​

App

​

Schema

• Fields: field name, type, and description.
• Overall extraction rules: natural language descriptions of overall rules.

​

Field

​

Quick Start

​

Step 1: Create an App

​

Step 2: Build a Schema

1. Generate a schema from a prompt
2. Generate a schema from smart suggestions
3. Build a schema from scratch

Detailed instructions and steps can be found in the corresponding section.

​

Step 3: Test and Call API

​

Out-of-the-Box Agents

• Zero Configuration: No prompt tuning or complex environment setup required. Experience true out-of-the-box functionality.
• Fully Automated Processing: Automated extraction begins immediately upon upload, delivering a seamless and fluid workflow.
• Comprehensive Document Coverage: Supports core business documents including Invoices, Receipts, and Purchase Orders (PO). Includes pre-configured extraction for over 15 key fields, such as document number, date, and amount.
• Built-in Expert Schemas: Comes with verified, high-precision extraction strategies designed specifically for business documents—enabling general-purpose models to achieve expert-level performance.
• Superior Accuracy & Performance: Extraction accuracy exceeds 90%, with a 33% boost in overall performance (processing time compressed to under 20 seconds per page). Both accuracy and speed meet industry-leading standards.

1. Access Ready-to-Use extraction capabilities for Invoices/receipts.

2. Select to upload an invoice/receipt document, only single document upload is supported.
3. The platform automatically classifies and extracts documents at the document level, and displays the results.
   1. Through a dropdown menu, users can manually change the automatically classified document type. After the type is modified, the platform will re-run the extraction task accordingly.
   2. Extraction results can be viewed in either a visual layout or JSON view, and can be exported in JSON or Excel format.

• Ready-to-Use capability supports integration with Human Review. When users call the Ready-to-Use capability via API, if review rules are enabled for Human Review, a Human Review task will be automatically created.
• Human Review applications associated with Out-of-the-Box capabilities do not currently support setting rules related to document confidence levels.

For more detailed instructions on human review operations, please refer to the human-in-the-loop workflow.

​

Configuring Schemas

• Generate a schema from a prompt – Use natural language to describe your extraction requirements, and ADP will automatically generate a schema.
• Generate a schema from smart suggestions – Let ADP analyze your document and propose intelligent, pre-configured instructions that you can refine.
• Build a schema from scratch – Manually define each field and rule with a structured form, giving you complete control over schema design.

​

Generate a schema from a prompt

1. Enter a very clear and detailed prompt. Specify the exact fields you want to extract. Explain how they are labeled in the document, and if you want them labeled differently in the extracted results.

Prompt best practices include:

• Prompt best practices include:
• Field names (e.g., Order Number, Customer Name, Contact Information)
• Field descriptions (e.g., “Transaction Amount should be numeric,” or “Delivery Date is the deadline requested by the customer”)
• Overall extraction rules (e.g., “Return company names in original language, translate minor languages, keep English and Traditional Chinese unchanged”)- Overall extraction rules (e.g., “Return company names in original language, translate minor languages, keep English and Traditional Chinese unchanged”)

2. Click Generate Now to create a schema instantly. You can fine-tune the output by editing fields or adjusting rules.

3. If you want to reset, click Back to Prompt Generation. This clears all configurations irreversibly and takes you back to the schema generation interface.
4. You can either save or test the schema once completing configuration.

​

Generate a schema from smart suggestions

1. Upload a document.
2. Select a prompt from “Try these Prompts”.

3. The schema will be generated automatically. You can fine-tune the output by editing fields or adjusting rules.

4. If you want to reset, click Back to Prompt Generation. This clears all configurations irreversibly and takes you back to the schema generation interface.
5. You can either save or test the schema once completing configuration.

​

Build a schema from scratch

1. Click + Custom Fields to start with a blank schema.

2. Add fields with name, type, and description.

• Supported field types include Text, Date, and Table.
• Field descriptions can include semantics, formatting rules, common document locations, or aliases.

3. Additionally, you can specify Overall Extraction Prompts—for example, handling missing values or post-processing extracted results (e.g., “Only capture numeric parts of Quantity, Price, and Amount, stripping currency symbols and units”).
4. If you want to reset, click Back to Prompt Generation. This clears all configurations irreversibly and takes you back to the schema generation interface.
5. You can either save or test the schema once completing configuration.

​

Version Control

• Supports saving extraction configurations as different versions, each with independent field configurations and overall extraction requirements.
• Meaningful names can be set for versions for easy identification and management.
• Supports viewing, renaming, or deleting historical versions of configurations.

• Extraction effects can be compared between different versions to select the optimal configuration.
• Supports continuing editing based on historical versions for rapid iterative optimization.
• Any specified version can be published to production environment for API calling.

1. After completing extraction rule settings on the configuration page, click the “Save” button to automatically generate the first version.

2. All saved versions can be viewed on the historical configuration page for comparison and management.

3. Select the version to be published and click “Publish” to complete deployment.

​

Schema Import and Export

• Supports exporting current extraction configuration as .config format files.
• Exported files contain complete data including field names, field descriptions, overall extraction requirements, etc.
• Facilitates configuration backup, team sharing, and version management.

• Supports importing configuration files exported from the ADP platform.
• Modifications and optimizations can be made based on imports.

1. On the application to be exported - Configuration - Historical Configuration page, select the configuration version to be exported, click [···] - [Export] button.

2. Download the generated .config configuration file.

3. On the target application - Configuration - Extraction Configuration page, click the [Import] button.

4. Upload the configuration file, and after successful upload, it will be automatically imported as a new configuration version.

​

Advanced Settings

Note：

• After modifying the configuration of the “mode” and “highlight”, if you choose [Save], the new configuration will not be applied to historical documents, but will only take effect on newly uploaded documents after the configuration change.
• After modifying the “mode” configuration, if you choose [Test], it will refresh the current preview effect, re-recognize the current document, and deduct the corresponding credits.
• When “highlight” is disabled, back-annotation will not appear in parsing results, extraction results, or human-in-the-loop collaboration results.
• All three parsing modes (Standard, Advanced, Agentic) consume the same amount of credits.

1. In Schema, turn on the “Long Document Extraction ” switch.

2. Enter 【Document Type】 and 【Document Feature】
   1. Supports inputting up to 5 document types
   2. Inputting page ranges is not supported.
   3. Writting document features can refer to best prompt practices.

3. The platform automatically locates the specified range for precise extraction and displays the extraction results.
   1. Clicking on the page number label supports source document highlighting and displays the document type to which the page number range belongs.

2. The extraction results can be switched between visual view and JSON view, support export in JSON, EXCEL, and Markdown formats, and support specifying the page range for export.

​

Viewing Extraction Results

​

Testing Schemas

1. Click Test at any stage of configuration to instantly preview how your rules perform on the current document.

2. Extraction results are displayed in two modes: a Visual View, which highlights extracted values directly in context, and a JSON View, which provides structured machine-readable output.

3. Results can be exported in Markdown or JSON format, or you can copy individual field outputs with a single click.

4. Once the configuration meets your requirements, click Save to publish the schema into production. The updated configuration is immediately reflected in API calls.

​

Confidence Scores

• Confidence score ranges from 0-1, with higher values indicating stronger model confidence in the field extraction result.
• Confidence is positively correlated with recognition accuracy - the higher the confidence, the higher the probability of accurate recognition.
• Supports setting confidence thresholds for different fields to trigger manual review (see Human-in-the-Loop Collaboration section).

• Quick problem identification: Focus on low-confidence fields for manual review and optimization.
• Review rule triggering: Automatic review rules can be configured based on confidence thresholds.
• Effect evaluation: Overall extraction configuration effects can be evaluated through confidence distribution.

​

Viewing Parsing Results

​

Text Type Labels

• Annotations: Notes and explanatory text in documents
• Formulas: Mathematical formulas, calculation expressions
• Body Text: Main content paragraphs of documents
• Footers: Information at the bottom of pages, such as page numbers, copyright information
• Headers: Information at the top of pages, such as titles, chapter names
• Images: Visual elements such as pictures, charts
• Chapter Titles: Document chapter division titles
• Titles: Various levels of headings and subheadings
• Tables: Structured data tables

• Assists model in understanding document structure, improving extraction accuracy of complex layout documents
• Users can intuitively view text types of each section on the results page, facilitating analysis of model recognition effects and layout features
• Supports intelligent filtering and retrieval based on text types

​

Parsing Result Export

• Complete document parsing results, including text type annotations.
• Structured content display maintaining original document hierarchy.
• Supports inclusion of complex elements such as tables, titles.

• Direct reuse of parsing results for downstream processing.
• Building enterprise knowledge bases and digital asset centers.
• Data integration with external systems.

​

Source Document Highlighting

• Automatically locate and highlight corresponding extraction fields in the original document.
• Support clicking extraction results to quickly jump to original text positions.
• Provide intuitive visual correspondence display.

• Verify the accuracy of extraction results.
• Locate specific positions of extraction errors.
• Analyze model extraction logic for configuration optimization.

​

Visual Correction of Extraction Results

• After users upload documents, the platform automatically parses and generates extraction results based on the latest configuration version.

• Supports users validating model extraction results on the extraction results page and manually modifying expected results.

• Visual display of correspondence between extraction values and original text.

1. View model auto-generated extraction values on the extraction results page.
2. Manually correct inaccurate extraction results.
3. System records correction information for subsequent batch evaluation and model optimization.
4. Supports batch completion of validation operations to improve processing efficiency.

​

Batch Evaluation

1. Click “Initiate Evaluation,” select the configuration version to evaluate, and initiate batch evaluation of multiple documents based on the specified version to quickly assess whether model extraction performance meets launch criteria.

2. After batch evaluation tasks are completed, the platform automatically generates offline evaluation report files supporting user download and viewing. Reports include overall evaluation metrics (accuracy, recall, F1 value, etc.), field evaluation results, document evaluation results, and field extraction details, helping users quantify model extraction performance at multiple levels. Field extraction details show comparisons between extracted values and expected values for each document, facilitating error sample location.

​

Out-of-the-Box Agents

​

Overview

• Zero Configuration: No prompt tuning or complex environment setup required. Experience true out-of-the-box functionality.
• Fully Automated Processing: Automated extraction begins immediately upon document upload, delivering a seamless and fluid workflow.
• Comprehensive Document Coverage: Supports core business documents including Invoices, Receipts, and Purchase Orders (PO), with 15+ pre-configured key fields such as document number, date, and amount. Also covers 11 common ID/certificate types including business licenses, bank cards, account opening permits, Chinese passports, ID cards, and driver’s licenses.
• Built-in Expert Schemas: Comes with verified, high-precision extraction strategies designed specifically for business documents — enabling general-purpose models to achieve expert-level performance.
• Superior Accuracy & Performance: Extraction accuracy exceeds 90%+. Simple documents such as ID cards are processed in 1–5 seconds per page; complex documents are processed in under 20 seconds per page. Both accuracy and speed meet industry-leading standards.

1. Entry point: Out-of-the-Box → Document Parsing / Invoice & Receipt
2. Select the desired Out-of-the-Box capability and upload the corresponding document. Only single document upload is supported.

Note:

• Out-of-the-Box extraction supports integration with Human Review. When calling the Out-of-the-Box capability via API, if review rules are enabled in Human Review, a review task will be automatically created.
• Human Review applications linked to Out-of-the-Box capabilities do not currently support setting rules related to document confidence levels.
• For more details on Human Review operations, see the Human-in-the-Loop Workflow section.

​

Invoice / Receipt

​

Use Cases

• Accounts payable (AP) automation for cross-border e-commerce
• Multi-supplier, multi-language, multi-template invoice entry
• Shared Services Center (SSC) operations
• Tax compliance and audit

​

Getting Classification and Extraction Results

1. Entry point: Out-of-the-Box → Invoice / Receipt

2. Upload an invoice or receipt document. The platform will automatically classify and extract the document.

​

Predefined Field List

​

Invoice

​

Receipt

​

Custom Extraction Fields

• Predefined Fields: System built-in default extraction fields (e.g., Invoice Number, Total Amount), ready to use out of the box.
• Custom Fields: Fields you add based on your specific business requirements (e.g., PO Number, Payment Terms).

1. On the Invoice/Receipt detail page, click the Field Configuration entry to open the field configuration panel.

2. The panel displays all active fields, including field name, description, and data type. Predefined and custom fields are distinguished by labels, with predefined fields listed first.
3. If a predefined field does not apply to your business scenario, click the Delete button next to that field to remove it.

4. Click the Add Field button and fill in the following:

Note: Each app supports up to 50 custom fields.

5. Click Save / Test after completing the configuration. The new field will take effect on the next extraction.

Note: Self-optimization samples collected before a field configuration change will no longer count toward the optimization cycle.

6. To revert to the system default configuration, click the Restore Default Fields button.

Important: Restoring defaults will delete all custom fields, restore predefined fields, and invalidate any applied self-optimization results.

​

Self-Optimization Agent

1. Correction Collection: When you edit and save a field value on the extraction result page, the system automatically records the correction.
2. Optimization Analysis: After accumulating corrections from 5 documents, you can start optimization. The Agent analyzes error causes and develops improved extraction strategies.
3. Auto Verification: The system automatically compares old vs. new extraction strategies using sample documents.
4. User Confirmation: Once verified, the optimization suggestion is presented for your approval.

1. On the Invoice/Receipt detail page, review the extraction results.
2. If a field value is incorrect, click directly on the field value to edit it.
3. Save your changes. The system will automatically record this correction as a learning sample in the background.

Note: Only corrections where the modified value differs from the original extraction are collected as learning samples. Normal editing operations are not affected.

4. Once 5 or more corrections have been accumulated, the Start Optimization button in the Self-Optimization Agent panel on the detail page becomes active.
5. Click Start Optimization. The system will analyze error patterns in the documents and generate an optimization proposal.

Note: Optimization analysis runs in the background and typically takes a few minutes. It does not block your other operations.

6. If verification passes, the Self-Optimization Agent panel displays the optimization suggestion, including:
   1. Change Summary: Describes the optimization results.
   2. Target Field Results: Original accuracy vs. optimized accuracy (e.g., total_amount: 60% → 93%).
   3. Impact on Other Fields: Shows whether other fields are affected.

7. You can choose:
   • Apply Optimization: Confirm and apply. Future extractions will use the optimized strategy.
   • Dismiss: Reject the suggestion. The current extraction strategy remains unchanged.
   • Re-optimize: Reject the suggestion and re-analyze to define a new extraction strategy.

8. If verification does not pass, the Self-Optimization Agent panel displays a notification. Click “Got it” and choose to either re-optimize or continue adding correction samples to improve optimization quality.

Tip: More samples lead to more accurate optimization by the Agent.

9. After applying an optimization, the system returns to learning mode. As you continue using and correcting, you can trigger further rounds of optimization to continuously improve extraction accuracy.

Note: Each optimization round uses all accumulated correction data to ensure globally optimal results.

10. In the Self-Optimization Agent panel on the detail page, you can view it at any time:
    • Number of accumulated correction samples per field
    • Current optimization status (Learning / Optimizing / Pending Confirmation)
    • History of applied optimizations

• All optimized extraction strategies revert to their initial state
• Accumulated correction sample data is preserved (not deleted)
• The optimization round counter is reset

Important: This operation shares the same entry point as “Restore Default Fields” for custom fields. Restoring defaults resets both field configuration and optimization strategies.

​

Purchase Order

​

Use Cases

• Supply chain procurement automation; bulk PO entry into ERP/SAP
• Three-way match for accounts payable (PO – Goods Receipt – Invoice)
• Supplier reconciliation and fulfillment tracking
• Standardized processing of purchase orders across regions and templates

​

Predefined Field List

​

Agentic Docflow

​

Overview

​

Key Concepts

Agentic Docflow is designed for compound tasks that follow a sequence such as “first do A, then do B, and finally output C.” For example, you can split bidding documents first and then extract fields by document type, or identify contracts and invoices first and then check whether the amount, supplier, and date match.

​

Use Cases

• Finance review: Check whether the amount, supplier, and date are consistent across contracts, invoices, OA forms, and other documents.
• Receipt processing: Split multiple invoices or receipts and extract information such as receipt number, date, amount, currency, and supplier.
• Document sorting: Identify and split mixed document packages containing business licenses, certificates, contracts, authorization letters, and other materials.
• Order review: Check whether orders are abnormal based on product information, unit price, validity period, and other fields.
• Financial report analysis: Extract financial report data, organize key metrics, and generate change analysis.

​

Quick Start

​

Step 1: Create an Agentic Docflow

• On the Home page, enter your document processing requirement in the Agentic Docflow input box and upload the documents to process.
• In the left navigation bar, go to My Apps, click Create App, and select Agentic Docflow.

Check whether the amount, supplier, and date match across these contracts and invoices.

Classify and organize the business licenses, certificates, and contracts in this bidding document, then extract key information.

I have a batch of orders. Check whether the product information, unit price, and validity period look abnormal.

Analyze this financial report, extract all table data, and highlight three numbers worth paying attention to.

​

Step 2: Refine the Docflow Through Conversation

• Add fields to extract.
• Describe how different document types should be identified.
• Specify the final output format.
• Add consistency checks, anomaly rules, or result consolidation logic.

To generate a more accurate docflow, include the document types, processing steps, fields to extract or validate, and the expected final output in your prompt.

​

Step 3: Test and Publish the Docflow

​

Create and Edit Docflows

​

Create a Docflow with Natural Language

I have a batch of orders. Check whether the product information, unit price, and validity period look abnormal. Output the abnormal order list in Markdown and explain the reason for each anomaly. The fields to extract and detailed review rules are included in the attachment.

Process this order.

The clearer your prompt is, the closer the generated docflow will be to your actual business requirement. We recommend specifying what the input is, what needs to be checked, and what the final output should look like. For complex long-running docflowa, you can build them in stages. For example, if a docflow needs to first classify and filter a batch of files by filename, and then extract and review the filtered files, you can first describe the classification and filtering requirement. After the assistant builds that part and the test result looks good, describe the extraction and review requirements to complete the docflow.

​

Adjust a Docflow Through Conversation

• Add or remove extraction fields.
• Modify document classification types.
• Add review rules or anomaly detection rules.
• Change the final output format.
• Ask the assistant to regenerate the configuration of a specific node.

Add an "Anomaly Reason" column to the result, and indicate whether the issue is caused by an amount mismatch, supplier mismatch, or date mismatch.

The bidding document may also include authorization letters. Split them into a separate document type and extract the signing date and authorization period.

​

Test a Docflow

​

Run a Test

1. Click Test Run.
2. Upload the documents you want to test.
3. Click Start Run. The docflow will start running in the background. Hover over Run Records to view the status of all historical runs.
4. Click Run Records, then click the file name to open the run detail page. On the detail page, you can view the execution status of each node and the final output from the End node. If the result does not meet expectations, continue refining the docflow through conversation or node configuration.

​

Adjust Test Results

• Incorrect document splitting: add document types, type characteristics, page ranges, or other identification rules.
• Inaccurate field extraction: add field definitions, common locations, aliases, and format requirements.
• Incorrect review results: add business rules and anomaly handling logic.
• Unexpected output format: specify whether the output should be a table, JSON, Markdown, or a summary report.

For complex scenarios, we recommend testing with 1-2 representative sample documents first, then validating stability with a larger batch of documents.

​

Publish and Use Docflows

​

Publish a Docflow

​

Use a Published Docflow

​

Prompting Best Practices

I need to process [business scenario]. The input documents include [document types].
First [step 1], then [step 2], and finally [step 3].
The fields to extract or validate include [field list].
The final output should be [output format], with the following special rules: [rules].

​

Example 1: Contract and Invoice Consistency Check

I need to process a reimbursement review scenario. The input includes contracts and invoices, which may be uploaded as one or multiple files.
The workflow should:
1. Identify the document type from the uploaded files, including contracts and invoices.
2. Extract the buyer, seller, and total amount including tax from invoices.
3. Extract Party A, Party B, and total amount including tax from contracts.
Finally, check whether these fields are consistent and output a review report. If any values are inconsistent, indicate the mismatched fields and the corresponding files.

​

Example 2: Bidding Document Classification and Extraction

I need to process a bidding document package. Companies may scan all materials into one large PDF, ranging from dozens to hundreds of pages. The materials may include but are not limited to bid letters, legal representative authorization letters, business licenses, performance records, ownership structure forms, bid summary sheets, bid bond payment vouchers, and service commitment letters.

The workflow should:
1. Automatically identify and split document types such as business licenses, certificates, contracts, and authorization letters. Clearly provide the page range for each material type.
2. Extract fields from each material type based on the fixed templates below.
3. Consolidate and output structured results by document type.

Fields to extract:
   (1) From the bid letter:
    - Bidder name, bid validity period, whether it is a consortium, whether it is an agent
    
   (2) From the legal representative authorization letter:
    - Authorization effective date, ID validity period, social insurance certificate period
    
   (3) From the business license:
    - Company name, unified social credit code, business term
    
   (4) From the performance record table (this is a table field):
    - Performance record table [No., contract name, contract amount]
    
   (5) From the ownership structure form (this is a table field):
    - Ownership structure form [Name of natural person/entity investing in or controlling the bidder, investment ratio (%)]
    
   (6) From the bid summary sheet:
    - Total quotation, tax rate
    
   (7) From the bid bond payment voucher:
    - Bid bond payer, bid bond amount
    
   (8) From the service commitment letter:
    - Delivery period, warranty period

​

Example 3: Financial Report Analysis

I need to analyze a financial report. The input is a PDF annual report, usually 100-200 pages.
Extract all table data, including the balance sheet, cash flow statement, and income statement, and organize the output based on the fields I provide in the attachment.
Finally, highlight three numbers worth paying attention to and explain why they may need further review.

Notes:
1. If the annual report contains data for multiple years, such as 2023 and 2022 shown side by side, extract the data separately by year.
2. The final output should be a standard Excel table. Each field should include four columns: value, source, calculation formula, and year.
3. Extract only data from the original document. Do not include units in the extracted results.

​

Notes

• Agentic Docflow is best suited for multi-step, cross-document tasks that require judgment or consolidation. If you only need to extract fixed fields, you can continue using Agentic Document Extraction apps.
• Docflow generation quality depends on prompt clarity and sample document quality. We recommend providing clear business rules and representative test documents.
• Before publishing, complete test validation to avoid using unverified configurations in production.
• For long documents, scanned files, and mixed document packages, specify document types, identification characteristics, and output requirements in the prompt.

​

Human-in-the-loop Workflow

​

Key Concepts

• Human-Machine Precise Routing, Ensuring 100% Business Accuracy: High-confidence and compliant data passes directly, low-confidence or non-compliant data is precisely routed to human interface, with final business data accuracy approaching 100%.
• Obtain High-Quality Ground Truth, Enabling Model to “Get Smarter with Use”: Each manual correction is recorded as high-quality annotation data, and the system generates configuration optimization suggestions based on ground truth data, continuously improving model performance.
• ROI Quantification Dimension Upgrade, from “Single-Point Accuracy” to “End-to-End Automation Rate”: No longer focusing only on technical metrics, but measuring core business metrics such as end-to-end automation rate and manual review rate, with customer automation rates increasing from 72% to 82% within one month.

​

Configuring Review Rules

• “Amount greater than 5000”
• “Invoice number is empty”
• “Delivery date earlier than current date”

1. Enter the Human Review module, click “Add Rule” or “Configure Rule,” and select the published application to which review rules are to be added.
2. Describe review conditions in natural language in the rule input box.

3. The platform automatically structures, parses, and generates rules based on LLM capabilities.
4. When saving rules, validity is automatically verified. After confirming all rules are correct, click “Confirm” to save.

​

Task Management and Processing

​

Task Management Center

• The task list page clearly displays the status of all tasks, such as “Pending,” “Approved,” “Cancelled,” etc., helping reviewers understand task processing progress at any time.
• Displays key information such as task creation time, task number, and rule name.
• Quick operation buttons for task processing, viewing results, canceling or deleting tasks improve task management flexibility.

​

Task Processing Interface

• Fields hitting review rules are highlighted, and the model’s reason for triggering review is displayed.

• Click “Previous” and “Next” to quickly switch between pending tasks.
• Support one-click “Approve” or “Cancel” task operations.
• Provide field correction functionality supporting direct editing of incorrect extraction results.

1. Enter the human-in-the-loop module.
2. View the pending task list, select the task to be processed, and click “Process Immediately.”
3. View highlighted problem fields and trigger reasons.
4. Confirm whether extraction results are correct. After completing result validation, click “Approve.” If the task does not require processing, click “Cancel.”

​

Credits Guide

​

What Are Credits?

• Credits are Laiye ADP’s universal billing unit.
• Each action (parse, extract, human review) consumes a fixed amount of credits per page.

​

How Credits Are Consumed

• Credits are deducted only for successful operations. If a task fails due to a system error, no credits will be charged.

Note: When your account runs out of credits, all document processing and API services will be automatically suspended. Please monitor your balance to ensure uninterrupted service.

​

Credit Consumption Rates

Note:

• Credit consumption will be affected by factors such as document type and model configuration.
• The credit consumption of Human Review depends on the base capability module it is combined with (such as Document Extraction App, purchase orders, Invoices). When used alone, Human Review consumes** 1 credit**. For better understanding, please refer to the following typical scenarios:
  • Scenario A:** Automatic extraction of Invoices only**
    • Calculation: 1.5 (base) =** 1.5 credits per page**
  • Scenario B: Purchase Order extraction with Human Review enabled
    • Calculation: 1.5 (base) + 1 (Human Review) = 2.5 credits per page
  • Scenario C: Document Extraction App with Human Review enabled
    • Calculation: 1 (base) + 1 (Human Review) = 2 credits per page

​

Consumption Example

1. Process 30 two-page documents using the Document Extraction App.
2. Extract a 12-page section from one 200-page document using Long Document Extraction.
3. Process 10 two-page documents with Human Review.

1. Document Extraction: 30 documents × 2 pages/doc × 1 credit/page = 60 credits
2. Long Document Extraction: 1 document × 200 pages × 1 credits + 1 document × 12 pages × 0.5 credits = 206 credits
3. Human Review: 10 documents × 2 pages/doc × 2 credits/page = 40 credits
4. Total Consumption: 306 credits

​

Credit Types and Validity

Note： If you want to obtain more credits, please contact the official staff of Laiye to recharge.

​

Checking Your Balance

• Quick View: Your current credit balance is visible in the user profile section at the bottom-left of the platform.
• Detailed History: For a full breakdown of your credit consumption, past orders, and transaction records, navigate to the “Account Management” section.

​

(注：内容由 AI 生成，请谨慎参考）