You can now clearly see the number of already vectorized documents and new attachments linked to each scraping job.[scraping-ui]
Web page vectorization now flows through a configurable workflow, making the process more flexible and controllable.[ingestion]
You can temporarily disable the new web ingestion workflow using a safety toggle if needed.[config]
A preconfigured workflow for web page ingestion has been added, optimized for data quality and reliability.[ingestion]
After ingestion, web page data is automatically enriched and finalized to ensure completeness.[canvas]
Components handling web page sources and sectors have been simplified for greater stability and clarity.[canvas]
You can now upload web pages directly in HTML format for processing, yielding more accurate results.[ingestion-bridge]
When you say "No" or "Rephrase" in chat, the system understands you want to correct your request and responds accordingly.[chat]
Aggregate queries over documents by topic now work correctly, returning more accurate results.[canvas-gate]
You can now filter and analyze candidates by birth date in recruitment management, with historical data automatically updated.[hr]
Clarification responses are now more contextual and reliable, thanks to advanced rules and a document metadata registry.[chat]
You can now select the "AUTO" option to assign roles automatically and re-vectorize many documents in one operation.[ingestion-ui]
Extracted fields are now automatically tailored based on sector and sub-sector, improving accuracy.[extractor/preset_fields]
If you provide a custom schema, the system uses it to extract data instead of falling back to a default.[bridge/preset_fields]
You can now pass dynamic schemas and sub-sectors to the extractor, making data extraction more adaptable to your documents.[bridge/models]
Metadata schemas can now be automatically resolved based on the selected sector and sub-sector.sector-definitions
Added a new 'selezione-recruiting' sub-sector to improve HR document management.sectors/hr
Optional sub-sector definitions can now be added to YAML configuration files for greater flexibility.sectors
The memory pipeline is now natively integrated into the canvas interface, improving reliability and performance.eval
Evaluation interface tabs now have consistent widths with a modernized, polished design using shadcn 4+.eval-ui
Added the INSURANCE sector with optimized reference data and a more accurate progress counter.eval
You can now assign different roles to individual files during bulk ingestion for finer control.ingestion
System admins can now enable or disable specific features for each target company.auth
Document search is now more accurate with a new algorithm combining multiple relevance criteria (BM25F-style).rag
Documents are now analyzed with title-aware embeddings and legal-context awareness, reducing clarification loops.rag
Overall RAG quality has been enhanced through a comprehensive end-to-end review and improvement.rag
All admins can now filter canvases by purpose, with access to all seven available purpose types.canvas-list
Ingestion nodes in the canvas now display clear schemas for metadata, inputs, and outputs, making configuration more intuitive.canvas
Your subscription-enabled features are now exposed via the auth API, enabling simpler integrations.auth
Ingestion experience improved with richer Layer 0/1/2 metadata, role-aware Rivettorizza, a streamlined bulk wizard, and auto-refreshing drawers.ingestion+monitor
Document chunks now preserve detailed metadata such as provenance, bounding box, and chapter path, enabling more precise search results.ingestion
The jobs sidebar now displays DSL-based ingestion jobs alongside existing BullMQ jobs for a unified view.jobs-sidebar
System canvases with corrupted domain presets are now automatically repaired to ensure proper functionality.scripts
Document IDs are now automatically pulled from trigger payload data when available, reducing upload errors.ingestion-source-upload
Ingestion jobs can now access trigger data during execution to customize behavior dynamically.ingestion-engine
The job execution context now includes trigger payload data for greater flexibility in processing.ingestion-types
Webpage content in Qdrant is now updated with sectorType and AI category tags for improved classification.scripts
Webpages missing AI categories are now automatically filled in, with database syntax fixes for improved stability.scripts
Every document upload now automatically creates a DSL-based ingestion job, with an optional legacy fallback controlled by environment.ingestion
A clear SourceBadge now appears in the monitor view to indicate the origin of each document.monitor
You can now view and manage collections on a dedicated list page with a reusable detail drawer.raccolte
You can now edit sector and visibility settings directly in the review queue, with a bulk-edit toolbar for multiple items.review-queue
The review queue now includes search, three preset filters, and the ability to filter by ingestion job ID.review-queue
In bulk ingestion Step 3, you can create jobs and view real-time progress of ongoing processes.ingestion-hub
In bulk ingestion Step 2, you can select an ingestion profile and toggle custom path rules for finer control.ingestion-hub
You can now select and upload multiple files at once using a card view and by dragging files directly into the interface.[ingestion-hub]
A smart banner highlights available products—you can dismiss it once you no longer need it.[ingestion-hub]
You can easily switch between single and bulk upload modes, and your preference will be saved for next time.[ingestion-hub]
URLs have been simplified: you now go directly to /documents/upload to upload documents, with no confusion.[routing]
All document ingestion features are now grouped under a dedicated 'Ingestion' section for clearer navigation.[sidebar]
You can update multiple documents at once, and changes are automatically synced with the AI search system.[documents]
Each document now shows an AI confidence score, indicating how reliable the automatic data extraction is.[ai-metadata]
You can now easily label the source of documents using a standardized field, improving traceability.[documents]
AI tools are now better organized internally for consistency and easier maintenance.[canvas-tools]
The AI remembers your previous preferences and interactions to provide more personalized responses.[canvas/prompt-context]
You can allow the AI to read and write personal information to remember your context across conversations.[canvas-tools]
The AI can now safely read content from a specific URL, with built-in protections to prevent security risks.[canvas-tools]
The AI can now extract text from images (like screenshots or scanned documents) with high accuracy.[canvas-tools]
The AI can now perform mathematical calculations and know the current date for more accurate and up-to-date responses.[canvas-tools]
The AI can now search for documents, read their metadata, and identify key topics to respond in a more contextual way.[canvas-tools]
You can now connect and use external tools directly from your workflows through a unified language.canvas-tools
You can now control how external tools connect to data sources, improving the reliability of responses.canvas-tools
Tool calling is now simpler and more integrated, allowing models to use tools seamlessly during conversations.llm-service
vLLM models now support non-streaming tool calling, improving performance for fast and precise responses.gpu
Google Gemini now supports tool calling in both streaming and non-streaming modes, for greater flexibility.custom-llm
Anthropic now supports streaming tool calling, enabling more responsive and fluid responses.custom-llm
Anthropic now also supports non-streaming tool calling, for scenarios requiring greater stability.custom-llm
Streaming calls with OpenAI are now more consistent and easier to integrate with the rest of the platform.custom-llm
OpenAI now supports non-streaming tool calling, offering a reliable option for complete responses.custom-llm
Chat messages and generation options have been expanded to better support tool usage during conversations.llm
Document ingestion workflows are now automatically converted into well-organized visual diagrams.canvas
System administrators can now filter features by purpose (All, Chat, Ingestion), improving management.canvas
A new document upload wizard and dedicated ingestion jobs panel make the process more intuitive.ingestion-dsl
CLI scripts are now available to easily migrate and rollback data.ingestion-dsl
Document uploads now use an intelligent routing system to direct data to the correct versions, ensuring compatibility.ingestion-dsl
Document migrations can now be paused and resumed at any time, with readiness checks and safe transitions to replace data without interruptions.ingestion-dsl
Every night, the system automatically compares new documents with existing ones to ensure quality and consistency, without disrupting daily operations.ingestion-dsl
You can set acceptable thresholds for document comparisons: if results exceed these limits, the system automatically alerts you.ingestion-dsl
You can create a data snapshot before major changes and easily roll back if needed, ensuring your documents are always safe.ingestion-dsl
The system tracks the status of every migration, allowing you to resume exactly where you left off—even after an unexpected stop.ingestion-dsl
The system automatically decides whether to route requests to old or new data, using predefined rules to maximize reliability.ingestion-dsl
Each company can choose the best migration mode and view detailed reports on comparisons and operation status.ingestion-dsl
All key pipeline components are now tested together, ensuring they work correctly as a single system.ingestion-dsl
Data is now processed in an optimized order to ensure information is always complete and accurate when saved.ingestion-dsl
The document library is now unified and role-based: filter and act on documents according to your role with clear, intuitive actions.ingestion-dsl
Uploading documents is now simpler: a 3-step wizard automatically detects file types and supports users with multiple roles.ingestion-dsl
A new dashboard clearly shows which features are active and which need configuration, with visual completion indicators.ingestion-dsl
Production data is now handled in shadow mode, with secure isolation per company to prevent cross-customer conflicts.ingestion-dsl
See the real-time status of ongoing processes with automatic updates and improved accessibility for all users.ingestion-dsl
Record management is now more reliable, as the system automatically handles document indexing in the background.ingestion-dsl
You can now filter and easily view ingestion components using a visual palette and an impact tray to understand the effect of your choices.ingestion-dsl
A new frontend API client lets you manage ingestion products directly from the user interface, simplifying configuration.ingestion-dsl
The document indexer now offers four ways to clean up outdated data and automatically orders content consistently.ingestion-dsl
You can now update or create document metadata and assign multiple roles to a single document, improving permission management.ingestion-dsl
You can now replace all operational data for a document in a single step, without needing to delete it manually.ingestion-dsl
A new service manages document indexing with greater precision and speed, built on advanced technologies.ingestion-dsl
The ingestion interface is now available in Italian, English, and Spanish, making it more accessible globally.ingestion-dsl
Each user role now has five pre-built ingestion templates ready to use, enabling quick workflow setup.ingestion-dsl
Four new activatable ingestion products have been added, expanding options for integrating different document types.ingestion-dsl
The ingestion wizard now automatically detects your role and suggests the most suitable components.ingestion-dsl
Text processing is faster thanks to simultaneous document processing using optimized hardware.ingestion-dsl
Advanced plan components have been added, along with a test framework to ensure proper functionality across all roles.ingestion-dsl
The system can now process multiple workflows simultaneously, improving performance with large document volumes.ingestion-dsl
Documents can now be sent simultaneously to two different ingestion paths, enabling multiple uses without duplication.ingestion-dsl
You can now choose which metadata to display or use for processing: all, only embedding-ready, LLM-ready, or none.ingestion-dsl
You can now classify roles in documents using keyword-based scoring and assign multiple roles to the same document.ingestion-dsl
New tools to automatically extract key metadata such as rule severity, tabular data, titles, keywords, and outcomes from documents.ingestion-dsl
You can now use dynamic placeholders to insert variable values during metadata extraction, making processes more flexible.ingestion-dsl
Documents can now be split into more precise segments using atomic markers and severity levels to improve structure.ingestion-dsl
Document examples can now automatically propagate outcome and sector metadata to generated segments.ingestion-dsl
The system now automatically detects and intelligently handles placeholders in document formats during chunking.ingestion-dsl
Operational documents are now split into rows and summaries, making it easier to analyze structured data.ingestion-dsl
Truth-based document chunking now uses domain presets and an advanced text-splitting algorithm for better accuracy.ingestion-dsl
14 preconfigured domain presets have been added to automatically optimize document processing for different industries.ingestion-dsl
A new advanced text-splitting algorithm has been integrated with guaranteed tests for improved reliability.ingestion-dsl
Domain presets are now centrally managed, making it easier to add or modify them.ingestion-dsl
You can now easily start the ingestion service using Docker Compose with a simple Make command.ingestion-bridge
Documents are now automatically routed to the most suitable parser based on their type (e.g., PDF, Word, etc.).ingestion-dsl
Unstructured documents (like PDFs or images) are now processed with higher precision using a dedicated external service.ingestion-dsl
Complex documents with advanced layouts are now analyzed using specialized technology to preserve their original structure.ingestion-dsl
Nodes can now send documents for processing, extract metadata, and check the ingestion service status.ingestion-dsl
Added a health endpoint to easily check if the ingestion service is active and running.ingestion-bridge
Documents are now automatically analyzed to extract title, keywords, outcomes, and key questions using advanced AI.ingestion-bridge
Documents are processed using two advanced technologies to convert PDFs and other formats into structured text.ingestion-bridge
Documents are split into consistent, uniform sections regardless of their original format.ingestion-bridge
Documents are split into consistent, uniform sections regardless of their original format.ingestion-bridge
Requests and responses between components are now more reliable and well-defined to prevent errors.ingestion-bridge
The ingestion service is now more stable and reproducible with a controlled environment and fixed dependency versions.ingestion-bridge
You can create, monitor, and follow document ingestion jobs in real time.ingestion-dsl
The system automatically loads four basic ingestion components, ready to be configured.ingestion-dsl
You can now build complete ingestion pipelines: from document upload to final storage.ingestion-dsl
All ingestion components are now centralized and easily manageable within the system.ingestion-dsl
Ingestion jobs are executed in the correct order and can resume from where they were interrupted.ingestion-dsl
Real-time progress notifications for ingestion jobs are automatically sent to users.ingestion-dsl
If the system crashes unexpectedly, it saves the job state and automatically restores it upon restart.ingestion-dsl
Ingestion jobs can now be resumed from where they left off, without needing to restart from the beginning.ingestion-dsl
Core data models and types have been added to support the new ingestion engine.ingestion-dsl
Each ingestion node now has a clear lifecycle and can run in different modes to suit workflow needs.ingestion-dsl
New data types have been introduced to better represent documents and events during ingestion.ingestion-dsl
You can now easily see and configure the limits and labels of connections between nodes in your workflow.canvas-schema
A new command-line tool checks your workflows for correctness and common errors.canvas
Draft workflows are now more strictly validated to prevent errors before saving.canvas-save
Stricter checks now prevent invalid connections between workflow nodes.canvas-validation
Nodes that use AI to generate text now have a more intuitive form to configure prompts.canvas-forms
You can now customize the role options available in nodes to match your specific needs.canvas-forms
Each categorization node can now use a different AI provider and model.canvas/categorize
Follow-up nodes can now independently choose which AI provider and model to use.canvas/follow-up
Each planning node can now be configured with a different AI model for more accurate results.canvas/planner
Evaluation nodes can now use different AI models to match specific evaluation criteria.canvas/evaluator
Gate nodes can now select the most suitable AI model for each decision.canvas/gate
You can now select a different LLM provider and model for each rewrite node, tailoring responses to the specific needs of each step in your workflow.canvas/rewriter
Classification nodes can now use customized LLM providers and models, enabling more accurate and context-aware categorization.canvas/classifier
Seven utility nodes now support selecting the LLM provider and model, giving you more control over how data is processed.canvas-types
Utility nodes can now generate responses more simply and consistently thanks to a new built-in function.llm
Queria automatically detects available vLLM models for planning and writing, simplifying initial setup.config
Custom company prompts are now automatically added to chats, widgets, and web searches, ensuring consistency with your brand voice.canvas-seed
Company prompts defined in the bot knowledge editor are now automatically injected into canvas workflows.canvas
You must explicitly save the LLM provider, model, and role settings to prevent incomplete configurations.llm-vault-fe
LLM node configuration forms are now standardized, making them more intuitive to use.llm-vault-fe
The LLM configuration form now includes advanced fields for role, provider, and model, offering detailed control.llm-vault-fe
A new model selector with suggestions and tags makes it easier to find and select the right model.llm-vault-fe
You can now see your LLM provider status directly in the node inspector, helping you avoid disruptions.llm-vault-fe
The provider selector warns you if a provider is missing or unconfigured, guiding you to fix it.llm-vault-fe
You can now select the model role (e.g., assistant, expert) directly in the node form to optimize behavior.llm-vault-fe
LLM provider and model settings are now accessible directly from the Company Settings page under the AI tab.llm-vault-fe
You can now manage all your AI models in a dedicated panel, accessible directly from the AI tab.[llm-vault-fe]
If the main model is unavailable, the system automatically switches to Qwen to continue generating text and plans.[llm-vault-fe]
You can now clearly see the status of each AI provider (not configured, connected, or invalid) on the main screen.[llm-vault-fe]
Set up and test your AI providers in a dedicated panel, with one-click options to save or remove settings.[llm-vault-fe]
Enter your API keys securely: toggle visibility and test them instantly without saving.[llm-vault-fe]
Quickly switch the active AI model using a keyboard shortcut, without navigating through menus.[llm-vault-fe]
Each AI provider is now clearly identified by an icon, name, and host, making setup more intuitive.[llm-vault-fe]
A visual indicator instantly alerts you if a configuration is valid or needs attention.[llm-vault-fe]
The AI interface is now available in Italian, English, and Spanish, with support for additional languages coming soon.[llm-vault-fe]
The list of available models now refreshes automatically when needed, even if you haven't changed settings.[llm-vault-fe]
All AI model operations (add, edit, remove) are now faster and more reliable.[llm-vault-fe]
Automatic AI component creation is now more stable and thoroughly tested, reducing risks of errors during document processing.[canvas]
Communication with the server to manage models and configurations is now faster and more reliable.[llm-vault-fe]
Data between frontend and backend is now consistent, reducing configuration errors and improving stability.[llm-vault-fe]
AI summaries of web search results are now displayed in your preferred language, without requiring manual translation.[web-search]
Key reasoning steps (like planner and critic) are now clearly labeled to improve understanding, even in multiple languages.i18n
You can temporarily disable the answer critic with a simple toggle, useful for testing or special cases.answer-evaluator
You can now see a clear summary of the planning and critic steps that led to the final answer.reasoning-panel
Assistant messages now display a badge indicating whether they are based on verified sources.chat-ui
If an answer is partial or unverified, a visible badge will clearly indicate this to users.chat-ui
Conversation logs now include structured details about planning and critic steps, enabling deeper analysis.observability
The default web search pipeline has been updated to work better with the new architecture.canvas-seed
The default chat pipeline now uses planner and critic components for more accurate and well-reasoned answers.canvas-seed
When an answer is regenerated, the critic provides specific suggestions to improve it.canvas-llm
If the critic detects missing evidence, the system expands the search to find more information.query-rewriter
A new 'planner' component helps the system intelligently decide which documents to retrieve before answering.canvas
The system now clearly distinguishes between search failures and generation failures, enabling more targeted fixes.answer-evaluator
The system can re-execute critical steps with greater attention, but with safety limits to prevent infinite loops.canvas-engine
Subscription data and configuration settings have been automatically updated to support the new version.llm-vault
You can now precisely select the role, provider, and AI model for each step, giving you greater control.canvas
You can now access AI models through a secure REST API authenticated and restricted to your company only.llm-vault
AI models are now automatically registered and validated, with a cache that speeds up response times for up to one hour.llm-vault
Built-in support for major AI models: OpenAI, Anthropic, and Google—ready to use without complex setup.llm-vault
You can now create, edit, delete, and securely encrypt AI model data with built-in validation and testing.llm-vault
Each company can now securely configure its own AI providers with automatic migration of existing data.llm-vault
You can now use advanced operators like 'includes', 'length', and 'greater than' to filter and compare data in your documents.canvas-resolver
When creating a scraping job, you can now assign document type, topics, and schedule execution all in one step.scraping-jobs-v2
You can now filter jobs by parent job and easily see the types of attachments from extracted web pages.scraping-jobs-v2
In the WebPage detail view, you now see all attachments with a button to import them directly as documents.scraping-jobs-v2
Each extracted attachment now has an 'Import as Document' option to quickly add it to your document library.scraping-jobs-v2
The WebPages table now uses color-coded badges to instantly show the status of each page.scraping-jobs-v2
Status badges now clearly indicate whether a page is NEW, UPDATED, DELETED, ACTIVE, or FAILED.scraping-jobs-v2
Each job card now displays scheduling details and a button to view the execution history.scraping-jobs-v2
The execution history now includes child REFRESH jobs, giving you a complete view of all automated updates.scraping-jobs-v2
You can now precisely configure how often and in what mode a job should check for changes on websites.scraping-jobs-v2
You can now access a dedicated interface to manage all your scraping jobs directly from the Web Scraping page.[scraping-jobs-v2] feat(scraping-jobs-v2): JobsV2Panel mounted on WebScrapingPage
Each scraping job now displays a progress bar and lets you refresh or cancel it anytime.[scraping-jobs-v2] feat(scraping-jobs-v2): JobCard with progress bar + refresh/cancel actions
Create new scraping jobs with flexible options: choose the scan type, starting URL, depth, and maximum pages to collect.[scraping-jobs-v2] feat(scraping-jobs-v2): NewJobDialog (type/baseUrl/depth/maxPages/urlList)
The interface automatically adjusts: updates data more frequently when you're active, and less when idle, for optimal performance.[scraping-jobs-v2] feat(scraping-jobs-v2): adaptive polling hook (2s active / 15s idle / pause hidden)
The user interface now communicates faster and more reliably with the scraping system, improving responsiveness.[scraping-jobs-v2] feat(scraping-jobs-v2): frontend service client for /api/scraping/jobs
You can now import downloaded files directly as documents, which are automatically analyzed and made searchable.[scraping-jobs-v2] feat(scraping-jobs-v2): attachment import endpoint (download + Document + vectorize)
The system now supports new APIs to manage scraping jobs, making it easier to integrate with other tools.[scraping-jobs-v2] feat(scraping-jobs-v2): mount /api/scraping/jobs/* REST routes
New API endpoints allow you to create, modify, and monitor scraping jobs programmatically.[scraping-jobs-v2] feat(scraping-jobs-v2): REST controller for /api/scraping/jobs/*
Documents are automatically updated on a schedule, without requiring manual intervention.[scraping-jobs-v2] feat(scraping-jobs-v2): worker handles scheduled-refresh-tick by creating REFRESH child
You can now create, edit, delete, and schedule scraping job updates with just a few clicks.[scraping-jobs-v2] feat(scraping-jobs-v2): scrapingJobService CRUD + schedule + triggerRefresh
Choose from four scanning modes: full site crawl, sitemap-based, single page, or targeted refresh.[scraping-jobs-v2] feat(scraping-jobs-v2): worker + 4 runners (INITIAL_CRAWL/SITEMAP/SINGLE_PAGE/REFRESH)
The system automatically detects changes on websites and only alerts you to pages that have been updated.[scraping-jobs-v2] feat(scraping-jobs-v2): change-detection helpers (sitemap candidates + diff)
Scraping jobs now make smart decisions about which pages to download or update, saving time and resources.[scraping-jobs-v2] feat(scraping-jobs-v2): processBatch with decidePageAction + applyDecision
Data is streamed in real time during scanning, so you can see results as they are collected.[scraping-jobs-v2] feat(scraping-jobs-v2): deepCrawlStream async generator for streaming NDJSON
All jobs are queued with visible status and progress; you can hide completed jobs without permanently deleting them.[scraping-jobs-v2] feat(scraping-jobs-v2): scrapingQueue + status/progress/soft-delete helpers
Added a script to automatically backfill content and media hashes in existing data, improving change detection.scraping-jobs-v2
Now automatically extracts attachments (like PDFs or documents) from links discovered by Crawl4AI.scraping-jobs-v2
New helper functions to generate unique hashes for content, media, and attachments, making it easy to detect changes.scraping-jobs-v2
Improved scraping job tracking with new data to trace content origins and detect changes.scraping-jobs-v2
Automatically imports external services defined in an OpenAPI file, simplifying configuration.external-tools
Imports and updates external services from OpenAPI, removing outdated ones to keep configurations clean.external-tools
Support for importing external services defined with the latest OpenAPI version (3.1).external-tools
Execute external REST calls with timeout, authentication, and custom response mapping.external-tools
A centralized registry to manage all connected external services, more reliable and easier to use.external-tools
Choose the authentication type for external services: none, API key, or Bearer token.external-tools
Transform external service responses using a simple syntax to extract only the useful data.external-tools
Automatically configure external call parameters (path, query, header, body) with various encoding styles.external-tools
Standardizes the format of external service responses and errors for more consistent handling.external-tools
Added data models to reliably store external services and their operations.external-tools
Sources at the bottom of responses now show the exact page number where each piece of content was extracted.citations
Citations now show individual document chunks instead of aggregated documents, for greater precision and traceability.citations
Page numbers are now displayed in source headers, making it easier to locate content in the original documents.citations
The citation pipeline has been improved: works in streaming, renumbers correctly, uses consistent colors, and highlights document regions.citations
User identity is now clearer in timelines and chats, with enhanced memory to prevent context loss.chat,timeline,memory
System administrators can now easily select which pipelines to use for attaching or detaching conversations.canvas
You can now duplicate system default configurations and lock them to prevent accidental changes.canvas
Conversations are now more reliably bound to pipelines, with a more robust UI for system settings.canvas
A new component displays additional message metadata, such as source and context, for greater transparency.canvas
Four new adaptive default pipelines have been added to better respond to different types of questions.canvas
A new evaluator checks whether answers are well-grounded in documents and complete, improving reliability.canvas
A fast filter blocks irrelevant responses before they are generated, saving time and resources.canvas
User questions are automatically rewritten using five different strategies, preserving the original meaning.canvas
A classifier detects the intent of questions among 8 categories, enabling more targeted responses.canvas
Conversation state is now persistently saved, allowing users to resume discussions at any time.canvas
New adaptive pipeline component types have been introduced, automatically adjusting based on context.canvas
You can now view format and AI-generated metadata directly in the document information popup.ui
Each document chunk now includes a format header to improve the accuracy of AI responses.rag
Queria automatically reads metadata from DOCX and PPTX files, such as author and modification date, for better organization.rag
Email files (EML) now provide complete metadata, such as sender, recipient, and date, for richer context.rag
A unified system has been created to manage and apply format metadata across all document types.rag
Thinking Block titles are now automatically generated from live reasoning content.chat
All reasoning responses now use the company's selected language for consistency and clarity.canvas
Thinking Blocks now follow ChatGPT’s standard reasoning pattern, with clear, logical steps for every response.chat
Activity labels in Thinking Blocks now cross-fade smoothly, like in ChatGPT, for a more natural experience.canvas
You can now choose how to display AI reasoning: detailed, simplified, or hidden, directly from the LLM form.canvas
Reasoning signals from vLLM are now automatically converted into readable tags in Queria.canvas
A dedicated model with the '/think' prefix is now used for reasoning to ensure more structured results.canvas
With the 'rawThinking' option, you can view the AI's original reasoning without any modifications or filtering.canvas
During company creation, the 'GENERAL' preset is now pre-selected for a faster start.rag
A 'GENERAL' sector has been added to evaluate AI performance on generic, non-industry-specific use cases.eval
Documents are now enriched with contextual information during processing, improving answer accuracy.[rag]
The GENERAL preset now uses an advanced system that processes text more intelligently and quickly.[rag]
Queria automatically detects the document type and selects the best extraction method for each case.[rag]
Documents are enriched with hypothetical questions and contextual headers to better understand their meaning.[rag]
Text is split at the sentence level for more precise and natural understanding.[rag]
Reasoning step labels are now available in Italian, English, and Spanish.[i18n]
You can now see Queria's reasoning steps directly in chat, as if they were part of the message.[chat]
You can now choose between GENERAL and CUSTOM presets via automatic metadata extraction from documents.[rag]
A new universal GENERAL extractor combines multiple methods to deliver better results on any document.[rag]
Queria automatically improves extraction of key data, regardless of the document format.[rag]
GENERAL is now the default setting, offering better performance for most users.[rag]
Queria's responses now show step-by-step reasoning, similar to Claude, for greater transparency.[chat]
You can now see how long Queria takes to reason and how many steps it goes through before answering.[chat]
Responses are displayed in real time as Queria reasons, so you don't have to wait for the full process.[chat]
Advanced reasoning mode is now available only to users with the appropriate subscription.[canvas]
Engine metadata is now saved directly in canvas-native messages, improving response traceability and consistency.[canvas]
Reasoning steps now include narrative labels that clearly explain the decision-making process.[canvas]
You can intercept and customize the AI's thinking stream during processing for finer control over results.[canvas]
Reasoning steps are now streamed in real time as events, letting you see the AI's thought process as it unfolds.[canvas]
You can now choose the depth of the AI's reasoning—from quick to detailed—based on your needs.[canvas]
Each purpose now has custom default settings, and inactive canvases appear dimmed for better visual organization.[canvas]
You can define purpose-specific default pipelines to automate your most common workflows.[canvas]
Answers in evaluation results are now rendered with Markdown formatting for clearer, more professional presentation.[eval]
Each question now saves its state automatically, you can delete unwanted runs, and the interface is more stable.[eval]
You can now monitor evaluation progress in real time and stop them with a single click if needed.[eval]
You can now directly select the topic and canvas to evaluate, without manual setup each time.[eval]
Evaluation start parameters now include unique topic and canvas IDs for more precise integration.[eval]
You can now start evaluations from the command line by specifying the topic and canvas—ideal for automation.[eval]
Canvases are now automatically resolved within the chat pipeline, improving response consistency.[eval]
All internal components now support topic and canvas parameters, ensuring a consistent flow from UI to backend.[eval]
Added a preset template for Hybrid Conversation Pipeline, combining automated responses and human interactions for complex use cases.canvas
All 23 missing components are now documented in AI Builder instructions, making it easier to create custom workflows.canvas
Added a new Smart Chat with Disambiguation template that clarifies users' ambiguous questions in real time.canvas
The query_router and intent_disambiguator components are now included in AI Builder instructions to better understand user requests.canvas
The logic for suspending and resuming conversations is now more flexible, enabling smoother handling of complex interactions.canvas
Added a new component to disambiguate user intents, improving response accuracy.canvas
The query router now shows confidence scores and possible alternatives, helping users understand why a specific response was selected.canvas
The query router now uses probability-based confidence scores to evaluate response quality.canvas
Introduced a lightweight default preset configuration for faster, simpler pipelines, ideal for basic use cases.presets
All 22 new components are now available in English, Italian, and Spanish, improving global accessibility.canvas
All 41 component types are now visible and accessible in the visual palette, simplifying workflow building.canvas
The 6 previously empty components are now fully functional, with real services performing actual operations.canvas
General improvements to pipelines, search page, and widgets, with full internationalization updates.
You can now customize the text guiding chart filters directly from the editor, without needing code.analytics
Citations now highlight the exact original phrases they were extracted from, improving traceability.citations
Start and end phrases are now properly propagated in the frontend to improve citation accuracy.citations
Document previews are now available for all file types, not just PDF and Office formats.citations
Start and end phrases are sent to the API to precisely locate their positions within documents.citations
The page-image endpoint now accepts parameters to highlight specific phrases in document images.citations
New feature to precisely locate the bounding boxes of start and end phrases in documents.citations
Start and end phrases are now linked to each citation to ensure consistency across the application.citations
Start and end phrase fields are now part of the structured citation schema.citations
A toast notification confirms when a workflow is set as the default.canvas
The 'Set as Default' button updates the default workflow for the entire company.canvas
Edit with AI opens the builder panel with the node's context pre-filled for faster editing.canvas
Right-click menu now lets you add, edit, duplicate, or delete nodes, including using AI to build them.canvas
Analytics results are now displayed directly in the canvas builder's output panel.canvas
Analytics results are extracted in real time during workflow processing.canvas
Analytics results are streamed in real time to the frontend via SSE.canvas
The analytics component is now registered and available for use in any workflow.canvas
Added a new visual component to analyze and display AI pipeline data in real time.canvas
Enabled the analytics component to be used within AI pipelines through a unified interface.canvas
AI Builder now applies changes directly to the canvas, eliminating the need for text-based editing.canvas
The AI Builder input field automatically expands as you type, making it easier to create complex instructions.canvas
Create and modify AI pipelines using simple natural language, without needing code or complex configurations.canvas
You can now filter and extract sector-specific metadata to improve the relevance of search results.canvas
The interface has been redesigned with expandable nodes, a tabbed palette, and a timeline trace to follow pipeline flows.canvas
A new two-step wizard guides you through pipeline creation, with badges indicating the main purpose of each flow.canvas
The creation wizard and purpose badges are now ready for translation into multiple languages.canvas
When creating a pipeline, the purpose is automatically assigned based on your choices, simplifying setup.canvas
Pipelines can now be categorized with predefined purposes and linked to specific business contexts.schema
You can now perform web searches and receive AI-generated summaries in real time, without waiting for full results.search
Web searches are now integrated directly into AI pipelines, allowing online results to be used as input.search
The data retrieval component now includes the ability to search web pages as an information source.canvas
The search endpoint is now connected to canvas pipelines, enabling a seamless and automated workflow.search
Added a widget assistant that guides users in building AI flows, with an updated plan for a better experience.[canvas]
All 43 components now support Italian and English, making the interface fully translatable.[canvas]
Added 18 new components to expand RAG, flow control, integrations, and AI processing capabilities.[canvas]
New components to run code, use templates, and send HTTP requests, expanding automation possibilities.[canvas]
All components now use a declarative schema, making configuration more stable and predictable.[canvas]
Connections between components are validated in real time, with errors shown via intuitive red arrows.[canvas]
Errors in components are now clearly shown with highlighted paths, enabling faster troubleshooting.[canvas]
The new universal configuration form works with legacy components, ensuring seamless backward compatibility.[canvas]
A single configuration form for all components, with dynamic fields that adapt automatically.[canvas]
14 reusable form widgets that make the configuration interface more consistent and easier to use.[canvas]
Customizable display options and translated labels for each component, with a clear namespace to avoid conflicts.[canvas]
Improved frontend schema types and API hooks for smoother data and configuration management.[canvas]
Soft warnings for incompatible connections, helping users fix errors without blocking the flow.[canvas]
Errors can now branch into alternative paths, allowing users to handle issues without stopping the entire process.[canvas]
The begin component now follows the declarative schema, ensuring consistency with all other components.[canvas]
Components in the canvas can now be registered and managed via a dedicated API, making interface customization more flexible.canvas
Canvas components now support static schema declarations, simplifying interface configuration and validation.canvas
A declarative type system for schemas has been introduced, enabling clearer and safer definition of data structures.canvas
Conversations are now accessible via a public API, with all messages saved on the server for reliable access.
Non-PDF documents are now automatically converted to PDF and cached, with precise coordinate mapping for search and analysis.
The user interface now smoothly displays PDF previews generated by Gotenberg, with faster and more reliable loading.
The previous preview generation system has been replaced by Gotenberg, delivering more accurate results and broader format support.
A new Gotenberg service has been added to convert any document to PDF, improving compatibility and quality.
The Gotenberg converter replaces PaddleOCR, eliminating the need for dedicated storage space and simplifying the infrastructure.
You can now easily filter documents by group using a dropdown menu on the monitor page.
The document list API now supports filtering by groupId, enabling more precise integrations with other systems.
In the monitor table, the group name is now displayed under the document type for better context.
During text analysis, a unique reference is now calculated for each chunk, improving the accuracy of linked evidence.
You can now assign multiple documents to a group at once directly from the monitor page, saving time.
During PDF processing, a coordinate map is generated and used throughout the pipeline to enhance search and analysis.
Queria can now automatically extract text from scanned PDFs, even with low-quality images, using intelligent OCR with GLM-OCR as a fallback.
You can now assign or modify the group for many documents at once using a simple interface designed for bulk operations.
Text extracted from documents now preserves its exact position on the page, improving answer accuracy and reference precision.
Each document can now be associated with a group, and you can update groups for many documents in a single operation.
The document list now shows which group each document belongs to, and you can update groups for multiple documents with one click.
Answers based on documents now show exactly where in the original text the information was found, for greater transparency.
The bulk group assignment interface is now available in multiple languages, making it more accessible globally.
When uploading documents from a URL, you can now specify their group directly, with a cleaner and faster experience.
The system now automatically accepts and validates the group you specify during upload, preventing assignment errors.
While uploading documents, you can now select or create a group directly from the upload form, without leaving the page.
A new intuitive menu lets you select or create document groups with just a few clicks, simplifying organization.
The group selection menu is now available in multiple languages, providing a consistent experience worldwide.
You can now override a document's language setting independently of its content, ensuring more accurate search results.
Documents uploaded from collections or chat are now weighted differently in searches, improving result relevance based on their source.
When searching for information, Queria now also examines automatically attached documents to deliver more complete answers.
Uploaded documents are automatically linked to the Raccolta when the conversation has a documentGroupId.
Added support for EML and P7M file formats, allowing upload and viewing of emails and digitally signed documents.
When selecting a conversation, the associated document group is automatically restored; switchers are locked after messages to prevent accidental changes.
API endpoints for creating or updating conversations now accept the documentGroupId, enabling association with specific Raccolte.
The documentGroupId is permanently stored in conversations and cannot be modified after creation, ensuring consistency.
The user interface now displays and manages the documentGroupId, making the Raccolta association visible in every conversation.
The document group associated with the chat is saved and automatically restored when reopening the conversation.
Conversation types now include the documentGroupId for more precise context management.
Added the ability to delete and preview documents directly from the toolbar, with a separator for better organization.document-groups
In the Raccolte panel and chat banner, you can now delete and preview documents with a single click.document-groups
Each topic can now specify its own Canvas pipeline, overriding global settings for finer control.topics
Raccolte are now directly accessible from the main navigation bar, even for external users.document-groups
Admins can now enable or disable Raccolte functionality for each subscription using a dedicated toggle.subscription
General case files have been organized into 10 Raccolte to demonstrate the practical use of this feature.demo
Full chat integration with Raccolte: save modal, picker panel, and active banner showing the current document group.document-groups
Added a new 'Collections' page accessible from the sidebar, to organize and navigate document groups.document-groups
Display clear status indicators (active, inactive, etc.) for each collection with an intuitive badge.document-groups
Multilingual support for Collections: interface available in English, Italian, and Spanish.document-groups
New tools to easily interact with Collections directly from the user interface.document-groups
New workflow dedicated to Generali insurance, using document groups to deliver more accurate answers.canvas
Search systems now consider document groups to deliver more relevant and contextualized results.canvas
Collections are now fully integrated into the app with dedicated navigation paths.document-groups
Access to Collections is controlled by subscription plan and user roles for enhanced security.document-groups
Create, edit, delete, and convert conversations into Collections, with automatic deletion of related documents.document-groups
Improved stability and reliability of Collection features through defined data types.document-groups
Data structure updated to support document groups and enable them only for selected subscriptions.schema
New blocks in the toolbar to categorize documents, assign agents, and collect user input.canvas
Easily export and import your workflows as JSON files directly from the canvas list.canvas
Technical support for exporting and importing workflows in JSON format, for sharing or restoring them.canvas
The system remembers previous user preferences and choices to automatically personalize responses.canvas
Added a new canvas for mortgage pre-assessment, specifically designed for Bank demos.[canvas]
Templates can now be automatically assigned to companies, simplifying use for different teams.[canvas]
Introduced a system default pipeline, a SAP canvas, cloning functionality, and improved permission management.[canvas]
Added 6 new ready-to-use templates for key sectors: insurance, legal, food, chemical, pharmaceutical, and tax.[canvas]
Full Spanish guide to canvases, with 6 pages of detailed instructions.[docsite]
Full English guide to canvases, with 6 pages of detailed instructions.[docsite]
New panel to easily select the system default canvas or your company’s default canvas.[canvas-pipeline]
Italian guide explaining all 15 canvas components for optimal platform usage.[docsite]
The system now includes a pre-configured default canvas to start workflows immediately.[canvas-pipeline]
Canvases can now be executed directly within chat, delivering more contextual and seamless responses.[canvas-pipeline]
Improved real-time communication between the system and canvases for faster, more reliable results.[canvas-pipeline]
Canvases can now use more system variables to better customize workflows based on context.[canvas-pipeline]
New option to enable advanced intelligent search techniques, improving answer accuracy.[canvas-pipeline]
After each response, the system automatically suggests follow-up questions to explore topics further.[canvas-pipeline]
The documentation site now fully supports Italian, English, and Spanish with automatic, localized translations.[docsite]
You can now automatically break down complex questions into simpler parts to get more accurate answers.[canvas-pipeline]
Questions are automatically routed to the fastest path for quick answers, without sacrificing accuracy.[canvas-pipeline]
The internal structure of pipelines is now more stable and easier to customize, with built-in support for type-safe data.[canvas]
You can now build complex workflows with loops, aggregations, and multi-step calls—all in one visual interface.[canvas]
The Agent Builder is now complete: it executes multiple operations in parallel and includes 5 new advanced tools for intelligent workflows.[canvas]
The user interface has been updated with new visual tools, intuitive forms, and multilingual support for the new components.[canvas]
Add custom variables, loops, and data operations directly into your workflow using new visual building blocks.[canvas]
Documents processed in fast mode are now marked with a lightning bolt icon, so you can spot them instantly in chat.[core]
Component parameters are now more flexible and easier to configure, giving you finer control over your workflows.[canvas]
Fast-mode documents automatically include relevant context, improving answer accuracy without delays.[core]
Parallel workflows now run smarter, executing independent steps simultaneously for maximum speed.[canvas]
Documents are now analyzed and ready for use much faster during upload.[core]
Temporary documents are now optimized for quick linking and immediate processing.[core]
You can now upload multiple files at once using drag-and-drop, clip buttons, or chips—no need to add them one by one.[attachments]
A smoother upload experience with a full-screen drag zone and visual badges for each uploaded file.[attachments]
The multi-attachment UI is now available in Italian, English, and Spanish for better global accessibility.attachments
You can now upload multiple files at once and easily manage attachments with bulk delete and retrieve operations.attachments
Temporary attachments are now automatically sorted for a more intuitive display during upload.attachments
Canvas agents now operate within chat, improving document generation and overall user experience.
You can now toggle chat functionality for each canvas directly from the editor toolbar.canvas
Reorder your canvases by drag & drop, and easily identify those with chat enabled via a visual badge.canvas
Select Canvas agents directly from chat and view responses in real-time with action tracing.chat
The topic selector now includes a dedicated agents section, simplifying the choice between topics and intelligent agents.chat
Multiple chat agent types have been added, with service methods to manage them more flexibly.canvas
Canvas-related requests are now processed directly by the Canvas Engine during chat, for more accurate responses.chat
A new service connects chat and the Canvas engine, ensuring seamless and uninterrupted communication.canvas
Agents can now access interaction history to deliver more contextual and consistent responses.canvas
Admins and users can now list, reorder, and assign chat agents using dedicated new endpoints.canvas
You can now find canvases with chat enabled, update display order, and reassign users to agents.canvas
Chat agents now only show canvases the user has access to, ensuring security and privacy.canvas
Added new fields to control chat, topic, sorting, and user permissions in canvases, enabling more precise workflow management.canvas
New ready-to-use templates for Legal Compliance Checks and Research Agents to automate complex tasks with just a few clicks.canvas
Updated user interfaces with multilingual support (English, Italian, Spanish) for better global accessibility.canvas
You can now resume interrupted work sessions with updated configurations and data types for seamless continuity.canvas
New components to assign tasks to agents or collect user input, with the ability to pause and resume workflows.canvas
Added the ability to automatically categorize documents and connect them to other system tools for an integrated workflow.canvas
Defined new data types and real-time events to improve workflow responsiveness and consistency.canvas
New topic selector and variable helper in the retrieval form to find and use data faster.canvas
A visual helper that shows all available variables and lets you copy them with one click, simplifying the use of dynamic data.canvas
The Agent Builder interface is now available in English, Italian, and Spanish for international users.canvas
Admins can now enable or disable the Agent Builder for each subscription, controlling access to advanced features.canvas
Pre-configured setup scripts for RAG Chat and Smart Router templates for quick, error-free installation.canvas
A new visual editor to build custom agents by dragging and dropping components, no coding required.canvas
Frontend structure improved to support the visual editor with smooth and stable performance.canvas
The full backend has been updated to support all new Agent Builder features with optimized performance.canvas
Added a visual interface to build custom agents with 5 pre-built components, making it easier to create workflows without coding.canvas
Improved workflow reliability with real-time validation and intelligent variable handling in the visual builder.canvas
Enabled saving and restoring visually built agent projects for a seamless, uninterrupted workflow.prisma
New graph nodes to display hierarchical summaries and community reports, making document relationships clearer.graph
Admins can now recompute document communities to keep search results always up to date.rag
Documents are automatically grouped into similar communities, improving the relevance of answers to questions.rag
Integrated advanced technology to automatically identify groups of related documents.neo4j
Documents are now analyzed hierarchically to generate more accurate and contextual summaries.rag
New system that creates multi-level summaries, enabling quick understanding of long documents.rag
Faster searches thanks to optimized indexes for summary levels and document communities.rag
Tables in documents are now automatically converted into natural language descriptions for better understanding.rag
Table descriptions are generated efficiently at scale, improving answer quality.rag
Table descriptions now have double weight in search results, making the most relevant information more likely to appear.rag
Table descriptions are now searchable as regular text, improving discovery of tabular information.rag
Automatic reduction of duplicate entity descriptions, keeping only the most important and concise information.rag
You can now search for documents using high-level and low-level keywords for more precise and contextualized results.[rag]
Search results now display high-level and low-level keywords to help you better understand the context.[rag]
Relationships between concepts in documents are now stored to uncover hidden connections between information.[rag]
You can now search for relationships between concepts using keywords, uncovering connections that were previously hard to find.[rag]
The system automatically generates questions from documents to improve search and content understanding.[rag]
Important fields in documents are weighted more heavily, bringing more relevant results to the top.[rag]
Document metadata is enriched with automatically extracted key questions and answers from the text.[rag]
Generated questions are saved and searchable within documents, making it easier to find specific information.[rag]
The system automatically selects the most relevant context for each question, improving answer quality.[crag]
An advanced system evaluates context relevance with higher precision, even if it takes longer to process.[crag]
For faster responses, the system uses a simple score-based path to quickly select the best context.[crag]
The system can now use external tools to answer complex questions, such as calculating data or accessing databases.[fast-pipeline]
You will be visually warned if a document connection is missing, preventing incomplete results.[network]
The network status icon now shows a yellow warning when a document connection is unavailable.[network]
The interface now correctly recognizes and handles the missing connection state for greater reliability.[network]
Verifies network drives are connected before browsing directories, preventing errors during navigation.network
Automatically checks network drive connections before each scheduled sync.network
Integrates network drive checks into system health monitoring to detect issues before they occur.network
Detects disconnected drives and automatically reconnects them without user intervention.network
Adds a clear status to indicate when a network drive is missing, improving monitoring.network
Allows full regeneration of all document vectors to ensure maximum search accuracy.scripts
Extracts content contextually by sector, applying predefined tags and domain terms to improve search accuracy.metadata
Adds the option to re-vectorize multiple documents directly from the interface, with updated translations.frontend
Supports sparse vectors for more efficient and accurate search in complex textual documents.qdrant
Improves text extraction from images with a backup system and precise extraction of mathematical formulas.ocr
Tracks document sections, validates boundaries, and adds provenance metadata for greater transparency.chunking
Removes duplicates between text chunks and splits tables intelligently to preserve their structure.chunking
More accurately removes standard headers and footers, improving the quality of extracted text.chunking
Allows viewing detailed content of extracted text chunks in JSON format for deeper inspection.documents
Shows a clear message during source verification to inform users that the system is checking citation reliability.chat
AI-generated section titles are now included in the search data, improving the relevance of results.pipeline
Search results are now ranked higher based on section titles, document titles, and chunk types to surface the most relevant content first.retrieval
AI automatically extracts meaningful section titles from each text chunk, eliminating the need for manual labeling.metadata
An automated tool updates all existing documents to include the new section titles and metadata.scripts
When Queria shows evidence for its answers, it now includes section titles to help you better understand the context.evidence
Citations in reports now display the section title they come from, making them easier to verify.citations
Document sources now display both the section title and the original document title for greater transparency.sources
The search system has been updated to support new metadata fields, improving search speed and accuracy.qdrant
Metadata extraction now happens in batches per chunk, making the process faster and more reliable.pipeline
Metadata is now built in greater detail for each chunk, improving search result quality.vectorization
Structured data like dates and amounts are now recognized and handled more precisely to improve search accuracy.vectorization
The Editor Agent now detects and blocks errors in generated documents, while flagging warnings to improve quality.docgen
A new service automatically validates generated documents to ensure logical and semantic consistency.docgen
Before selecting sources, the system automatically generates search queries based on the document schema to find the most relevant content.docgen
AI interprets the document schema to generate intelligent search queries, making source selection faster and more accurate.docgen
Added new data types to improve the handling and display of issues and results during document generation.docgen
New specialized source preview renderers for Legal, Pharma, Chem, and AE documents to improve clarity and context.source-previews
Extracted the Food Source preview component for better reusability and modularity across the interface.source-previews
Introduced shared types and a unified export system to simplify integration of external source preview renderers.source-previews
All 27 document templates now explicitly require appropriate sources, ensuring more accurate and consistent outputs.docgen
Tabular and Timeline buttons have been replaced with a unified ToolsPopover for a cleaner, more intuitive interface.tools
A new contextual menu organizes analysis tools—Tabular, Timeline, and Document Generation—into a single access point.tools
Document generation category is now passed in real-time during chat, improving result relevance.tools
Skips automatic category detection when the category is already provided, reducing response times.tools
Admins can now enable or disable Tabular Review and Timeline features directly from the subscription interface.tools
Added new configuration flags to support enabling/disabling Tabular Review and Timeline features in the user interface.tools
Access to Tabular Review and Timeline features is now controlled by subscription plans for better license management.tools
Subscription plans now include explicit options to enable Tabular Review and Timeline, improving configuration flexibility.tools
Citations are now validated faster with an automatic background check, reducing wait times.fast-pipeline
Optimized the fast query path with improved routing, enhancing response speed without sacrificing accuracy.fast-pipeline
The system can now automatically combine multiple data sources and AI models to answer questions more intelligently and with custom settings.fast-pipeline
Answers are now more organized and reliable, with an automatic fallback if the system takes too long to respond.fast-pipeline
You can now request answers in specific formats (like tables or lists), and the system will generate them automatically without manual adjustments.fast-pipeline
Structured answers work with any AI model, regardless of the provider, ensuring consistent results across all use cases.fast-pipeline
New protection against credential sharing with two-factor authentication via Google Authenticator for enhanced security.security
A new central registry to define and manage enterprise data types, with integrated documentation to make usage easier for everyone.semantic-layer
Daily updates of legal documents are now faster and more reliable, reducing manual work for administrators.legal-sources
The subscription management and admin settings interface is now more intuitive and easier to use.frontend
Relationships between documents and concepts are now more accurate and faster to process, thanks to technical improvements to the graph system.graph
The system now better understands questions by integrating the data registry, making responses more relevant and context-aware.semantic-layer
Administrators can now easily create, edit, or delete enterprise sector definitions directly from the interface.semantic-layer
Document access rules are now managed from a single point, ensuring consistency and security across the entire organization.semantic-layer
Responses to recurring questions are now faster thanks to intelligent caching of frequently used data.semantic-layer
Enterprise sector information is now stored in the database, making it more flexible and easier to update without code changes.semantic-layer
Disconnected documents are now automatically linked, and text chunks are color-coded by their source type for clearer visualization.graph
Added edge labels, legend, zoom controls, and improved spacing for clearer graph visualization.graph
Added 30 preconfigured YAML seed files for industry sectors to simplify initial setup.semantic-layer
Added a YAML loader with automatic validation to ensure configuration data is correct and reliable.semantic-layer
Integrated YAML dependency and a structured data model to define enterprise sectors.semantic-layer
Enhanced node design with gradients, shadows, and pills for a professional, modern look.graph
The Reasoning Graph link is now visible only to users with an active enterprise subscription.nav
The Reasoning Graph button is now subscription-gated and opens in a modal window instead of a new tab.reasoning
Rewrote the Graph Explorer page to match the Search Page design, now opening in a modal.graph
Added a dedicated modal for exploring reasoning graphs, styled consistently with other Queria features.graph
Added Italian, English, and Spanish translations for the Graph Explorer interface.i18n
Now you can filter conversations by specific topic ID in the API for better contextual search.reasoning-graph
The conversations endpoint now accepts a topicId parameter to filter results by topic.reasoning-graph
The Graph Explorer is now integrated into the sidebar and Reasoning Panel for quick access.reasoning-graph
Added new components: ReasoningCanvas, NodeDetailPanel, and GraphExplorerPage for full reasoning graph experience.reasoning-graph
Added a frontend API service and a dedicated landing page for the Graph Explorer.reasoning-graph
Added a new backend module to explore and reason about relationships between information, making document analysis more intelligent.reasoning-graph
Added a subscription flag to enable or disable the reasoning relationships feature, controllable by administrators.reasoning-graph
New tools to visually explore links between documents and concepts, with the ability to dive deeper into contexts and relationships.graph-explorer
Documents and topics now automatically update the relationship graph when created, modified, or deleted.graph
Integrated with Neo4j to keep the relationship graph updated in real time.graph
You can now easily delete timeline entries using a dedicated delete button in the dropdown menu.timeline
See an animated progress bar showing processing steps as the system generates responses or relationships, for greater transparency.timeline
Information is now displayed in an interactive chronological sequence with clear, easy-to-navigate popups.timeline
Existing data is automatically updated to include newly extracted relationships from documents.edge-facts
Relationships are now handled with greater precision, allowing skipping unnecessary steps and tracking metadata.edge-facts
Relationships are automatically prepared after saving and cleanly removed when a document is deleted.edge-facts
New service to search, update, and manage relationships with access controls and activation/deactivation options.edge-facts
Relationships are now generated and formatted more clearly and contextually, with automatic quality checks.edge-facts
Personalized responses now consider your user profile, making results more relevant to you.memory
Instead of auto-injecting suggestions, the system now proposes them non-intrusively, letting you choose when to act.timeline
AI Memory now opens as a slide-in panel from the avatar dropdown, making it more accessible and intuitive.ui
An AI-powered system automatically detects and generates chronological sequences to clearly visualize document timelines.timeline
AI responses now include an inline timeline to clearly show the chronological order of referenced facts.chat
Extracted data cells update in real time during extraction, without requiring a page reload.tabular
Long cells are truncated for better readability, with a side panel to compare columns and updates every second.tabular
The search system now uses a more advanced model (Layer 1.5) to find the most relevant documents before responding.embedding-router
The system automatically builds an intelligent index from database logs to improve search speed and accuracy.embedding-router
You can now configure and customize the search system with new options and dedicated data directories.embedding-router
You can now ask for an automatic summary of a document’s content using a predefined question.meta
A new intelligent classification service improves result accuracy, with automated tests to ensure reliability.embedding-router
Analytics reports now include rich metadata, ideal for enterprise teams analyzing document usage.analytics
AI now builds its reasoning steps from a coherent narrative, making responses more logical and easier to follow.reasoning-v5
The UI now shows AI reasoning steps in real time, with a new panel and a toggle to enable or disable the feature.reasoning-v5
Editors can now view and restore deleted documents, filtered by topic, for more precise control.rbac
AI responses update in real time during generation, showing the narrative plan, steps, and summary as they are created.reasoning-v5
Chunk difference results are now logged to update only modified parts, making content processing faster and more efficient.vectorization
The reasoning system has been enhanced with new prompts and metadata to deliver more contextual and accurate responses.reasoning-v5
New schemas and types have been added to support more complex and customized reasoning.reasoning-v5
The system suggests follow-up questions based on previous interactions, making conversations more natural and intuitive.memory
Personal queries are processed faster thanks to an optimized path and a dedicated memory prompt.memory
Editors and admins can now access the dashboard directly from the user dropdown menu.ui
Admins can now view detailed usage metrics for memory across all workflows.memory
A new 4-layer memory system has been implemented to better manage information and improve response quality.memory
When documents are updated, only changed parts are re-processed, saving time and resources.documents
Each content chunk now has a unique identifier to quickly detect changes.vectorization
An automatic script has updated all existing documents with unique change identifiers.scripts
Web page chunks now include unique identifiers, preparing the system for future incremental updates.scraping
Specific modified chunks can now be removed individually, without reloading the entire document.qdrant
A new service automatically identifies changed text parts and updates them precisely.vectorization
Each document chunk now includes a unique field to track changes and optimize processing.schema
Users can now access multiple topics with controlled access roles, enhancing security and personalization of experiences.
Search weights are now automatically passed from the execution plan to the retrieval system, delivering more targeted results.orchestrator
A migration script simplifies the transition to sparse vectors without disrupting existing data.migration
When enabled, the system sends sparse vectors to Qdrant for more precise and efficient searches.vectorization
Sparse search combined with a new weighted fusion technique improves the quality and variety of results.retrieval
Searches now support named vectors, enabling simultaneous searches across different types of data.qdrant
A new search strategy combines three methods with custom weights for more reliable results.search
The execution plan now includes retrieval weights, providing greater transparency and control over results.planner
Adding and updating documents now supports named and sparse vectors for more flexible data management.qdrant
Data structure is now optimized for named and sparse vectors, enabled via a simple flag.qdrant
New GPU methods simplify processing of complex queries and documents, improving performance.gpu
Batch document processing now supports sparse vectors, making indexing of large data volumes faster.embedding
New data types and a dedicated method enable more efficient handling and generation of sparse vectors.embedding
You can now view the timeline of legal articles via a new endpoint, providing richer context.legal-sources
The system now retrieves legal article timelines directly from Neo4j, ensuring up-to-date and connected data.legal-sources
Effective dates and amendments to legal texts are now automatically synced to the database, ensuring up-to-date information.legal-sources
Abrogated legal documents are now clearly marked in search results, helping users identify which laws are no longer in effect.backend
Added technical tools to detect related groups of legal documents, improving the discovery of legal contexts.legal-sources
Legal search now includes articles from semantic communities, delivering more contextual and comprehensive results.backend
New functions to retrieve articles from a community and identify which community a document belongs to.backend
New endpoints to explore and search legal document communities directly from the interface.legal-sources
After each data update, the system automatically detects communities of related documents.legal-sources
Introduction of an advanced module to identify natural groups of similar legal documents.legal-sources
Neo4j-integrated methods to detect and manage legal document communities more efficiently.legal-sources
Service health checks are now grouped into collapsible sections, making the interface cleaner and easier to navigate.ui
A new intelligent filter evaluates query confidence before proceeding, reducing irrelevant results.cascading-router
An initial query filter removes clearly off-topic requests before any document search begins.cascading-router
Two new control layers added to ensure only the most relevant queries are processed, with tests for stability.cascading-router
Queries are rewritten for specific company contexts, and results are reordered more naturally and accurately.retrieval
The system optimizes ranking resources, showing the most relevant results faster and with less computational load.perf
Search is now faster thanks to a routing decision cache, a more compact prompt, and a Vespa-inspired adaptive retrieval system, plus a pipeline profiler to monitor performance.perf
The system now quickly returns relevant documents after initial processing, skipping unnecessary steps for faster responses.routing
The orchestrator now uses clear intents ('document inventory' and 'document search') to better understand user intent and deliver more accurate answers.routing
The system now explicitly recognizes whether the user is looking for a document inventory or specific information, improving answer relevance.routing
Web pages are now analyzed to extract key entities and preset metadata, enabling smarter and more consistent search results.
In the web page list and details, the RAG preset and subsector are now displayed for better context and organization.
When a topic is updated, associated metadata is automatically resynchronized, and documents needing refresh are flagged accordingly.
The UI now correctly displays the RAG preset and subsector for web pages and fixes an issue with topic assignment during document upload.
The system now formats numbers and currency according to Italian conventions (e.g., €1,000.50), improving familiarity for local users.
Documents and web pages that need reprocessing after metadata changes are now automatically flagged for update.
A new shared service ensures all components use consistent presets, eliminating inconsistencies in responses.
The system now supports additional formats like OCR images and derived exports, expanding compatibility with diverse document types.
Document processing now includes enhanced metadata for more accurate results, with fixes to sector pipelines and inline tabular review.
The evaluation area now lets you easily select companies, collapse the sidebar, and fixes validation checks for answer quality.eval
Tabular review, the evaluation results page, and the sources detail panel have been improved for clearer and more intuitive viewing.
Queria now generates high-quality documents with enriched schemas, integrated Word tables, and intelligent narratives that adapt to context.
27 new pre-built templates across 5 categories have been added, with the ability to require fields in nested objects for greater precision.