Release Notes¶

This page summarizes notable changes in CORA over time.

Release 1.1.13 (2026-06-05)¶

Key Highlights¶

Expanded thread and document navigation in the frontend, including a thread panel on user-document pages and direct document access from chat reference popups.
Refined version-comparison UX in the Versions tab with a persistent Compare Versions control and improved loading behavior.
Strengthened PCT comparison in LLM-connect with a verification-and-gap pipeline, hybrid search, and restored eCFR/KBA regulation handling.

Improved¶

Added a thread panel on the user-document detail page for managing conversations in document context.
Enabled navigation to referenced documents from chat grounding/reference popups.
Made the Compare Versions control persist while scrolling in the Versions tab and added a dedicated versions-list loading skeleton.
Enhanced the PCT comparison agent with verification retrieval, gap extraction, expanded multi-document aggregation metrics, and hybrid search tuning.
Improved eCFR regulation prompts and re-enabled eCFR/KBA source handlers in LLM-connect.

Fixed¶

Fixed blank screen when sorting notes by title in the backend.
Aligned thread continue-button styling with the updated Figma design on library and project thread views.

Known limitations¶

When hovering over multiple documents in the versions list of a UNECE document, the button "AI Consultation" stays highlighted in Safari browser.

Release 1.1.12 (2026-05-30)¶

Key Highlights¶

Introduced industry-based onboarding and product behavior across backend, frontend, and LLM-connect, including signup industry selection, industry-scoped document access, filters, and conversation starters.
Added configurable AI model selection in chat with aligned backend/LLM model naming and message metadata.
Expanded sorting, list refresh, and state-persistence improvements across library, projects, user documents, requirements, and notes.
Improved UNECE version workflows with enhanced resolution, composition-manifest hierarchy, and checkbox-based version selection.

Improved¶

Added industry management on signup with metadata support and industry-based document access control and filter configuration in the backend.
Exposed sort_by on list APIs and numeric sorting for text fields.
Enhanced UNECE document version resolution and content-fetch behavior in the backend.
Added industry selection and industry-based conversation starters in the frontend, plus a config-driven AI model selector in the chat bar.
Implemented sorting across library, projects, user documents, requirements, and notes, with list refresh after add/edit/delete operations.
Persisted Project and User Documents page state across navigation and added checkbox selection in the Versions tab for clearer compare workflows.
Enabled nested composition-manifest hierarchy in the Versions tab and chat against series amendments/revisions.
Added library filter chips for quick document-source filtering and updated US document metadata display.
Scoped LLM agent responses to the user's industry and updated model naming/default-model configuration in LLM-connect.
Improved regulation discovery for eCFR/KBA sources, US ingestion grounding with the database, UNECE series handling, and chat response quality.
Stored LLM model identifiers on chat messages for improved traceability.
Updated pharmacy-oriented conversation starters and LLM models endpoint integration in the frontend.

Fixed¶

Resolved send-button layout shrinking after sending chat messages.
Fixed lists not refreshing after add, edit, or delete actions.
Fixed grounding-source loading state applying to all buttons when only one was clicked.
Removed the ISO filter from backend filter configuration and refreshed KBA filter labels.

Known issues¶

None identified in this release window.

Release 1.1.11 (2026-04-20)¶

Key Highlights¶

Expanded and refined filter configuration across backend and frontend, including deeper KBA structure, filter/search persistence, and production-specific US filter handling.
Improved version-comparison and chat experience with better defaults, performance, and thread-management safeguards.
Refreshed product branding to AAI Innovations with updated frontend logo assets.

Improved¶

Updated filter labels for clearer naming and consistency across filter views.
Enhanced KBA filter configuration by making it selectable and adding detailed Type Approval subcategories, document types, and designations.
Added US document-filter support in filter configuration for controlled rollout scenarios.
Persisted library filters and search across reloads, and refetched filters/T&Cs on language switch.
Improved version comparison UX by pre-filling baseline/target selections, adding clear controls, preventing duplicate selection, and improving inline comparison performance.
Enhanced UI behavior with Grounding Sources loading indicators, preserved markdown newline rendering, chat-in-progress indicators, and enablement of rename-thread/edit-project chat actions.
Improved KBA document visibility in the frontend by showing domain and sub-domain in library and detail views, and enabled chat from the Versions tab with CELEX-only graph display refinements.

Fixed¶

Made captions optional in grounding-source rendering to avoid failures when caption metadata is absent.
Prevented automatic new-thread creation when deleting the active chat thread.
Disabled "remove from project" for project-context threads to prevent incorrect detach actions.
Added Escape-key modal close behavior and restored chat more-actions availability.

Release 1.1.10 (2026-03-31)¶

Key Highlights¶

Introduced and refined UNECE document version comparison, including YAML-based comparison output and improved HTML preprocessing.
Improved multilingual documentation handling for locale-aware navigation and link rewriting.

Improved¶

Added backend support for YAML-based UNECE version comparison output.
Updated version comparison formatting and resolved version mismatch handling.
Removed table-of-contents content from processed HTML inputs to improve comparison quality.
Enhanced documentation language support, including locale preservation in navigation and rewritten links for multi-language docs.

Fixed¶

Fixed version mismatch behavior in document version comparison flow.

Known issues¶

Version comparison output can still vary for edge-case source formatting in highly customized legacy UNECE HTML structures.

Release 1.1.9 (2026-03-06)¶

Key Highlights¶

Token usage tracking and balance enforcement.

Improved¶

Token usage tracking and balance enforcement for usage and quota management.

Fixed¶

None in this release window.

Known issues¶

None identified in this release window.

Release 1.1.8 (2026-03-04)¶

Key Highlights¶

Signup improvements with blacklisted emails and automatic Terms and Conditions acceptance on successful signup.
Public Terms and Conditions route and backend ReqIF export for project requirements.

Improved¶

Public route for Terms and Conditions.
Signup flow improvements and email blacklist handling.
Automatic Terms and Conditions acceptance on successful signup.
Backend ReqIF export for project requirements.

Fixed¶

None in this release window.

Known issues¶

In some filter views, labels can differ by language because locale-specific mappings are still being used.

Release 1.1.7 (2026-02-28)¶

Key Highlights¶

Provenance in grounding sources and improved handling of request cancellation in LLM connect.
Refresh of signed URLs for images and requirement PCT source from markers.

Improved¶

Added provenance to grounding sources and fixed user-document limited provenance handling.
Requirement PCT source is now derived from markers.
Image signed URLs are refreshed when expired.
Request handling for LLM connect service in case of request cancellation.
Release notes updated for 1.1.6 and 1.1.5.

Fixed¶

User document limited provenance behavior.
Expired image signed URLs now refresh correctly.

Known issues¶

Filter labels can still vary by locale in edge cases where translation mappings diverge.
In some filter views, labels can differ by language because locale-specific mappings are still being used.

Release 1.1.6 (2026-02-26)¶

Key Highlights¶

Introduced consolidated multilingual filter label support for en, de, zh, and fr.

Improved¶

Centralized filter label data into a unified structure for easier maintenance and future language expansion.
Updated filter management flow to consume the consolidated labels consistently across API responses.
Refreshed OpenAPI output to reflect the latest filter-label related API behavior.

Fixed¶

Standardized filter labels across locales by retiring legacy language-specific mappings.

Known issues¶

Requirement numbering may appear inconsistent when prefix formatting is expected.
Filter labels can still vary by locale in edge cases where translation mappings diverge.

Release 1.1.5 (2026-02-17)¶

Key Highlights¶

Added startup automation for Atlas search index creation and expanded document chunk metadata for chat experiences.

Improved¶

Added image metadata to document chunks and exposed it for downstream chat usage.
Added database indexes to the relations collection for better query performance.
Updated Terms and Conditions language selection to follow the profile interface language.
Enhanced request logging to include user ID from request headers for better traceability.
Updated GitHub Actions runner and workflow cleanup behavior to improve pipeline stability.

Fixed¶

Added duplicate project name validation in create, update, and patch endpoints.
Updated requirement numbering behavior to use a prefix-based format.
Improved filter-language handling while preserving dedicated translation JSON support.

Known issues¶

In some filter views, labels can differ by language because locale-specific mappings are still being used.

Release 1.1.4 (2026-02-05)¶

Key Highlights¶

Added project activity history tracking and strengthened thread/document validation for project-linked workflows.

Improved¶

Implemented backend project events tracking with expanded API and schema coverage.
Enhanced request logging middleware and included user identity details for better observability.
Refactored thread handling to consistently set document_id, version, and language metadata.
Restricted UNECE document retrieval to series versions.
Enabled chat behavior when Croatian (HR) language support is available.
Refreshed end-user documentation with expanded guides and visual assets.

Fixed¶

Added duplicate-entity protection for notes and requirements to improve data integrity.
Prevented duplicate document names per user in user document workflows.
Clarified edge-case handling in requirement title-base extraction logic.

Known issues¶

In rare cases, duplicate notes or requirements can still be created for the same logical entity.
Users may encounter duplicate document titles within their personal document lists.
Requirement title-base extraction may not handle all edge cases consistently.

Release 1.1.3 (2026-01-26)¶

Key Highlights¶

Expanded project/thread management APIs and improved document discovery with Atlas-powered listing and richer metadata.

Improved¶

Added endpoint support for retrieving threads by ID and fetching multiple projects in batch.
Added attach/detach document APIs for project-linked workflows with clearer attached-document naming.
Added prompt endpoints and improved document listings with returned date fields.
Implemented MongoDB Atlas Search support for document listing.
Added CELEX source URL support in document relation responses.

Fixed¶

Standardized project document references to use title instead of legacy document_name.
Improved note-tag filtering by project ID.
Cleaned publication date formatting and adjusted thread file-title handling for better readability.

Known issues¶

Some project responses still return legacy document-name fields instead of the unified title field.
Note-tag results are not always scoped correctly when filtering by project.
Publication date text and long thread file titles may appear inconsistently formatted.

Release 1.1.2 (2026-01-16)¶

Key Highlights¶

Improved chat grounding fidelity across document types and stabilized backend runtime/dependency workflows.

Improved¶

Expanded grounding source payloads with additional metadata (oj, URLs, hrefs, image captions) and multi-document support.
Added thread-level attached-document locking and additional document details in list responses.
Added relation filtering controls to exclude specific relation types in graph lookup.
Enabled automated build and delivery for end-user documentation.
Migrated dependency management from pip to uv and updated Docker/dependency setup for more reliable builds.

Fixed¶

Fixed requirement generation behavior for PCT and discovery flows.
Fixed enum filter expansion for multi-select top-level sources (EUR-Lex + UNECE).
Fixed long-running user document upload instability.

Known issues¶

Requirement generation can fail or behave inconsistently for PCT and discovery scenarios.
Multi-select source filtering (EUR-Lex + UNECE) may return incomplete source coverage.
Long-running user document uploads can fail during extended processing.

Release 1.1 (2026-01-05)¶

Key Highlights¶

Strengthened discovery/chat responses and improved thread behavior for project and non-project workflows.

Improved¶

Added support for returning threads without project assignment when project_id is null.
Enhanced discovery responses to include graph_lookup context.
Added/updated LLM-connect and user-document endpoint behavior for better document interaction.
Improved PDF URL behavior to open documents in a viewable mode.

Fixed¶

Fixed missing relation subdivisions in relation responses.
Fixed chunk-processing edge cases where headings were absent.

Known issues¶

Relation responses may omit certain subdivisions in some graph outputs.
Some chunk records without headings can cause inconsistent downstream rendering/processing.

Release 1.0 (2025-12-24)¶

Key Highlights¶

Established the core CORA backend foundation, including auth flows, chat/thread architecture, document workflows, and project collaboration primitives.

Improved¶

Introduced foundational auth and profile capabilities, including signup/SSO integrations and Terms & Conditions checks.
Added core chat/thread APIs (ask/invoke/stream, thread state/listing, unified thread listing, discovery summary flows).
Built core document workflows including search/filtering, relations, source/metadata handling, and document info endpoints.
Added user-document ingestion and chat integration with queueing, cleanup, and grounding-source adaptation.
Added requirements and notes capabilities, including linking notes with requirements and improved requirement endpoint behavior.
Established delivery and quality pipelines with CI/CD, workflow hardening, and incremental versioning.

Fixed¶

Addressed multiple source-linking, language/version attribute, and endpoint consistency issues across chat, documents, and traceability flows.

Known issues¶

Relation responses may miss certain subdivisions in some graph outputs.
Chunk records without headings can lead to inconsistent downstream rendering and processing.