ComPDF AI Versions

Découvrez les nouveautés dans chaque version, avec des détails sur les fonctionnalités, les améliorations et les problèmes résolus.

janv.
févr.
mars
avr.
mai
juin
juil.
août
sept.
oct.
nov.
déc.
ComPDF AI v2.6.323 oct. 2026
Fonctionnalités
  • Release Overview:
    • Added tenant switching, department trees, custom roles, and data permissions to improve enterprise collaboration and data isolation.
    • Added extraction confidence scores, risk levels, and automatic confirmation to improve result review efficiency.
    • Expanded template sample management and classification profiles for more flexible configuration and automatic classification.
    • Added a WorkFlow Recycle Bin with file restoration, permanent deletion, and retention-based cleanup.
    • Improved self-hosted deployment initialization, license management, resource quotas, and operation auditing.
  • Tenants, Organizations, and Permissions:
    • Multi-tenant deployments support linking an account to multiple tenants and switching between them. Roles, permissions, and business data are loaded for the selected tenant.
    • Added department trees with department creation, editing, moving, deletion, and member assignment.
    • Supports adding, linking, editing, enabling, disabling, and removing members, plus CSV bulk import with pre-validation and per-row feedback.
    • Added custom role creation, editing, copying, enabling, disabling, and deletion, with multiple roles per member.
    • Supports feature permissions and data scopes covering the current user, the current department and its descendants, specified departments and their descendants, or all data.
    • Unified data-scope checks for lists, details, downloads, exports, and modifications. Permissions refresh promptly after member, role, or department changes.
    • Improved temporary passwords, mandatory password changes at first login, and password resets. Resetting a password invalidates previous login credentials.
  • WorkFlow Extraction and Result Review:
    • Added file-level and field-level confidence scores for VisualExtract and LayoutExtract, with risk levels, scoring status, and deduction reasons.
    • Supports table-cell scoring, multi-page result scoring, and field location mapping.
    • Added automatic confirmation based on scoring completeness, confidence, risk, and field validation results.
    • Supports rescoring after result edits and manual confirmation. Scoring failures do not revoke completed manual confirmations.
    • Improved parsing-result archiving, queries, and downloads for both extraction modes, and added parsing-result editing.
    • Added a unified result format for multi-page fields, table cells, and location associations while retaining compatibility with existing result formats.
  • Templates, Samples, and Automatic Classification:
    • Each template supports up to 5 sample files, with sample listing, upload, deletion, and replacement.
    • Added template classification profiles and version management. Profiles are rebuilt after sample or classification-setting changes, with retries for failures.
    • Classification profiles are isolated by tenant, user, template, and version to prevent classification data from being mixed when shared templates are used.
    • Template testing and AI-assisted configuration support selecting a sample. Existing sample-selection rules apply when none is specified.
    • Improved sample information in template details and compatibility with missing samples, deleted samples, and historical single-sample data.
    • Improved associations between template fields, output settings, and export names to prevent duplicate columns or empty new columns after field renaming.
  • File Management and Recycle Bin:
    • Added a WorkFlow file Recycle Bin with queries, filters, restoration, and permanent deletion.
    • Supports bulk restoration, bulk permanent deletion, and asynchronous clearing of large batches, with task-status queries.
    • Supports automatic cleanup based on the retention period.
    • Permanent deletion handles associated results, scoring archives, and storage objects. Failed cleanup operations can be retried.
    • Recycle Bin operations enforce tenant and data permissions to prevent restoring or deleting another tenant's files.
  • Open API and Result Export:
    • Expanded template sample, classification configuration, parsing-result editing, and related task APIs, with improved permission checks for Service API Keys.
    • Preserves existing /api/v2 paths and default v2 parsing results. Use Accept-Version: v3 explicitly to retrieve simplified parsing results.
    • Supports result_format=unified for unified extraction results. Existing clients can continue using the original format.
    • Added custom JSON export configurations for different versions and target systems, including a default configuration setting.
    • Improved per-file and aggregated CSV, Excel, and JSON exports, including compatibility with historical templates and table results.
    • Extraction exports retain business data and existing formats without confidence scores, risk levels, deduction reasons, or other scoring metadata. Internal scoring data is retained.
  • Resources, Notifications, and Operation Logs:
    • Improved tenant resource summaries with used, maximum, and remaining quotas and validity periods, distinguishing limited, unlimited, and unconfigured states.
    • Improved quota reservation, settlement on success, and release on failure to reduce usage inconsistencies from concurrent processing and duplicate requests.
    • Adjusted asset expiration handling to preserve login, tenant switching, and historical data management. Validity and quota checks run before operations that consume assets.
    • Added confidence-related conditions to reminder rules and improved quota, expiration, and subscription notifications.
    • Unified queries, filters, and CSV exports for Python platform and Java WorkFlow operation logs.
    • Added login time, source IP, tenant, and department context, with standardized operation types and detail codes for multilingual integration.
    • Improved log redaction, CSV encoding, and timestamp representation for more readable and compatible audit data.
  • Self-Hosted Deployment and Licensing:
    • Improved first-install initialization of tenants, administrators, permissions, and default templates, with support for resuming interrupted installations.
    • Added offline signed-license verification for licensed modules, validity periods, and member limits.
    • Added license-status queries and renewal capabilities, with stronger signature verification, clock rollback detection, and invalid-license blocking.
    • Improved external model service configuration and connectivity, supporting reuse of existing parsing, extraction, and embedding services.
    • Added pre-deployment checks, backup, upgrade, and rollback tools, with corresponding configuration and operational documentation.
  • Improved compatibility when switching samples, testing templates, updating classification profiles, and reading historical results.
  • Improved error handling for email delivery, operation-log queries, and embedding model URL configuration.
Corrections
  • Fixed historical table results appearing empty in unified results.
  • Fixed aggregated export failures for older templates, export groups overwriting others with the same name, and inconsistent field mappings.
  • Addressed table scoring objects being written directly into export files, causing incorrect business values and column structures.
  • Fixed inconsistent permission or resource information for some accounts after tenant switches, role changes, and member-status changes.
ComPDF AI v2.6.2Version majeure20 août 2026
Fonctionnalités
  • Release Overview:
    • Added the WorkFlow intelligent document extraction workspace, including template-driven extraction, task management, result review, and export.
    • Added Service API Key, Webhook, and v2 Open API capabilities for secure third-party integration.
    • Added the message center and reminder rules with in-app, email, and Webhook notifications.
    • Added output conversion and validation for extracted fields, and improved organization permissions, subscription assets, and operation auditing.
  • WorkFlow Intelligent Document Extraction:
    • Supports two processing modes:
      • LayoutExtract: Performs layout parsing before field extraction, with traceability for layout and extraction results.
      • VisualExtract: Performs direct visual extraction for standard, single-page, and handwritten documents.
    • Supports batch upload, asynchronous processing, task and file status management, retry, cancellation, re-extraction, and classification adjustment.
    • Supports file history, file details, source preview, extraction-result editing, confirmation, and confirmation revocation.
    • Added a results workspace with search and filtering, configurable columns, exception markers, and structured result export.
    • Improved extraction lifecycle handling with normalized result formats, retry behavior, processing status, review status, and legacy-result compatibility.
  • Template and Output Configuration:
    • Supports organization and default template views, template status management, ordering, group-template associations, and resumable personal drafts.
    • Supports template sample upload, AI-assisted template configuration, Visual/Integrated extraction testing, and test-artifact download.
    • Added output configuration for text, number, date, and enumeration fields:
      • Formatting, normalization, replacement, regular-expression extraction, numeric units, date conversion, and enumeration conversion.
      • Export field-name configuration and duplicate-name validation.
      • Required, format, and comparison validation.
      • Field-level exception markers in results.
    • Improved template-deletion safeguards with permission validation, default-template protection, and explicit handling of associated files.
  • Open API and System Integration:
    • Aligned the WorkFlow v2 Open APIs with the PRD, covering files, templates, batches, tasks, extraction results, and assets.
    • Supports Service API Key authentication through X-API-Key, Api-Key, or Authorization: Bearer {key}.
    • Derives organization identity from the API Key for organization data isolation; Open APIs remain separate from browser login-session APIs.
    • Supports metadata pass-through during upload, processing, result queries, and Webhook callbacks.
    • Added public reminder-rule CRUD endpoints at /api/v2/rules, using public DTOs and ISO 8601 UTC timestamps; creation returns HTTP 201 and deletion returns HTTP 204.
    • Standardized error responses with stable error_code values while preserving existing HTTP status codes and business code values.
  • API Key and Webhook Management:
    • Supports API Key creation, listing, name filtering, editing, enable/disable, deletion, copying, expiry, usage records, encrypted storage, and verification.
    • API Key lists use masked previews and never return complete secrets.
    • Supports Webhook creation, listing, editing, enable/disable, deletion, signing-secret copying, encrypted storage, event subscription, delivery records, retries, and latest-response queries.
    • Supports batch and file lifecycle callbacks, as well as internal HMAC event authentication between Java WorkFlow and the Python platform.
    • Improved handling of historical Webhook records with undecryptable secrets so that one invalid record cannot interrupt other pending deliveries.
  • Reminder Rules and Message Center:
    • Supports reminder-rule creation, editing, copying, enable/disable, deletion, scheduled evaluation, and event-driven evaluation.
    • Supports template-field conditions, system-status conditions, AND/OR logic, and rule-name conflict validation.
    • Added system fields available for condition matching:
      • review_status: pending, confirmed.
      • processing_status: lifecycle states including parsing, classification, extraction, failure, cancellation, and completion.
    • Provides inferred status values for legacy WorkFlow events: extract_complete maps to pending review/extraction completed, while confirm_result maps to confirmed review/extraction completed.
    • file_name, template_name, and detail_url remain available as notification-template variables and message links; they are no longer exposed as system condition fields.
    • Supports in-app, email, and selected-Webhook delivery. Email recipients can be organization members, direct email addresses, or extracted-field bindings; addresses are split, validated, and deduplicated, while invalid values are skipped without blocking other channels.
    • Added email/Webhook delivery retries, delivery diagnostics, trigger records, event-level idempotency, and rule-level deduplication. A successfully updated rule resets its own deduplication records; deleting a rule removes its deduplication records.
    • The message center presents system messages, system announcements, and user reminder rules; rule messages are grouped by rule name for directory counts.
    • Added system notifications for asset quota, plan expiry, and subscription activation, and standardized platform-generated notification text in English.
  • Organization, Permissions, and Knowledge Base:
    • When an administrator invites a regular member, the new member inherits enabled business permissions; personnel-management permissions are not inherited automatically.
    • Added frontend-recognizable error codes for login failures, duplicate users, and related cases.
    • Reminder-rule recipients can include organization members; regular members can read and select existing organization Webhooks.
    • Supports organization-member queries for knowledge-base member configuration while preserving the existing knowledge-base membership query.
    • Restored runtime compatibility for legacy knowledge-base prompts that omit the {knowledge} placeholder.
  • Assets, Subscriptions, and Auditing:
    • Supports subscription asset allocation and asset-account state synchronization.
    • Supports asset-expiry scanning, low-quota alerts, and plan-expiry notifications.
    • Improved compatibility for expired legacy asset accounts and standardized subscription-expiry display.
    • Added operation logs for API Keys, Webhooks, reminder rules, templates, files, tasks, results, and exports.
    • Added WorkFlow audit-log persistence and queries. Logs are written in an independent transaction after the primary transaction commits, and write failures are recorded without affecting the primary business transaction.
  • Documentation and Release Package:
    • Updated the Chinese user guide, Open API reference, WorkFlow interface reference, error-code reference, and release-package documentation.
    • The release package includes Docker Compose, service configuration, Nginx configuration, initialization SQL, and API documentation.
    • The public Open API contains only the 23 formal APIs defined by the PRD; test and template helper endpoints are excluded from the public API reference.