What's New in Prisme.ai

​

Features

​

AI Builder

• AI Builder: Transmit file size to the upload function
• Block DialogBox: Send file size when user upload a file
• prismeai-runtime: New truncate function to truncate strings in automations
• prismeai-runtime: New lower and upper functions for string manipulation
• Set instruction: New clone option to choose if the new variable should be a clone (true) or a reference (false) to provided value
• Set instruction: New RUNTIME_SETS_CLONE_DEFAULT environment variable to globally disable default clone behaviour (brings performance improvements)
• prismeai-crawler: Increased default max length
• prismeai-crawler: Added additional MIME types to support more document formats

​

AI Knowledge (AIK)

• AIK: Allow to restrict agents from being shared (https://docs.prisme.ai/products/ai-store/sharing-agents#restrict-sharing)
• AIK: Added Deep Research Tool
• AIK: Added Reranking, filtering, contextual compression to the RAG pipeline as native tools
• AIK: Added carbon footprint analytics
• AIK: Allow webhooks to return a “tags” field for vector search

​

AI Insights (AII) & AI Store (AIS)

• AII: Better display for feedbacks list
• AIS: New design with more information for products home

​

AI Builder

• Builder: Ability to prune import a Workspace from Settings/Packaging view with checkbox to delete files not in uploaded archive
• AI Builder: Enhance workspace list retrieval
• AI Builder: On archive import, automatically remove automations/pages/imports index entries from old files
• AI Builder: Allow superAdmins to delete apps even if their source workspace doesn’t exist anymore
• Builder: Update workspaces list display, include owner name & labels

​

AI Builder

• Pages: Better display of KPI block with long values
• Pages: Loading state on DataTable block with loading attribute
• Automations: Add shareToken option in Form block upload options
• Automations: RichTextEditor uploads its images as private with token access
• prismeai-api-gateway: Support SMTP server configuration for emails
• prismeai-api-gateway: New allowAnonymousUsers security option to prevent anonymous user access to builders APIs
• prismeai-api-gateway: Better management of lowercase login SSO to avoid duplicate accounts
• Vector Store: Add new vectorIndexPrefix configuration parameter for tenant separation in shared databases
• Collections: Parse connection string to extract query parameters & manually pass them to MongoClient constructor

​

AI Collection (AIC)

• AIC: Added QueryTextToSQL instruction for advanced text-to-sql from text input

​

Bugfixes

​

AI Builder (AIB)

• AIB: Fix workspaces display when user has less than 20 workspaces
• AIB: Fix appInstance slug renaming

​

AI Knowledge (AIK)

• AIK: fixed an issue on Vector store when using Elastic Search (index.knn)
• AIK: Fix warning on saving agent if only maxContext is provided but not maxResponseTokens
• AIK: Fix displaying proper error message when vector database fails initializing project vector index (Redis or ES/OS) after uploading first document
• AIK: Added error and warning upon saving settings when models don’t have expected specifications or Agent’s values mismatch
• AIK: Fix KPI stats on conversations
• AIK: Fix document upload with POST /document API and multipart/form-data format
• AIK: Fix embeddings error handling, catch any case where no embeddings are returned
• AIK: Fix chunk size taking precedent over embedding models set maxContext when calling embedding API
• AIK: Use model max response size instead of project’s if model allowed size is smaller
• AIK: Fix chat-completion not using the “model” field from Knowledge Client.chat-completion
• Conversation Service: Fallback to use 2nd thread if the first is corrupted or not in conversation anymore
• Conversation Service: Add index to children field in Messages collection

​

AI Store (AIS)

• AIS: Fix assistants public status being lost when editing settings from AI Store
• AIS: Failed messages (due to LLM outage, invalid requests, etc.) are no longer removed from conversation history

​

AI Builder

• prismeai-console and prismeai-pages: INTERNAL_API_URL no longer crashes console and pages
• prismeai-console: Fix crash when products name is localized
• prismeai-runtime: Runtime contexts management refactored
• prismeai-runtime: Fix SSE stream sometimes missing the first chunk (e.g., for Gemini models)
• prismeai-runtime: Make schedules great again and introduce their own events for performance reasons
• prismeai-runtime: Fix infinite loop on workspaces.apps.configured event handling
• prismeai-runtime: Fix redis persisted user context overwriting user auth data
• Vector Store: Fix vector index deletion with OpenSearch
• Blocks: Set images upload as public for presentation Blocks (Header, Image, Hero…)
• Pages: Fix KPI block display
• prismeai-api-gateway: Azure SSO improvements
• prismeai-api-gateway: Fix email validation not being taken into account
• Block RichTextEditor: Fix losing caret position on Firefox/Win when page is in studio iframe
• prismeai-workspaces: When config update is made but content is the same, do not trigger an update

​

Other Improvements

• prismeai-console and prismeai-pages: Upgrade to latest npm package versions

​

Bugfixes and improvements :

• Enhanced Group ACLs: Added groupAcls user field with API update capabilities and runtime synchronization
• Platform Role Protection: Fixed issue where platformRole was lost when patching other user fields from non-super admin accounts
• User Context Synchronization: Resolved runtime user context losing platformRole & groupAcls when receiving updates without these fields
• User Management: Updated user.groupAcls handling from AI Governance user webhook with new groupAcls parameter in PrismeaiAPI.updateUser
• API Optimization: Removed redundant /contacts API calls from user webhook as all user details are now available in user context
• Emit Instructions: Execute broker sends from emit instructions inside parent thread to prevent blocking interactions during high CPU/blocking work
• Topic Management: Both createUserTopic and joinUserTopic emits now execute inside parent thread for improved performance
• Wait Instructions: Wait instruction emits also moved to parent thread for better I/O processing
• Event Loop Control: Runtime now gives control back to NodeJS event loop when ending automations to prevent single interaction starvation
• Async Support: Added async option to Emit instructions
• Streaming Optimization: Removed final AIS callback on streaming end, replaced with Knowledge client.knowledgeMessage event from AI Store on final SSE chunk
• Database Indexing: Added MongoDB index to apiKeys field in Teams collection for improved query performance

​

Enhanced Model Management & Analytics

• Improved Model Cards: Comprehensive model information including:
  • Model capabilities and specifications
  • Models organized by provider
  • Detailed descriptions and use cases
  • Cost information per model
  • Carbon footprint metrics for environmental impact tracking
• Message-Level Statistics: Individual AI agents now display detailed stats per message:
  • Processing duration
  • CO2 emissions per interaction (when configured)
  • Cost breakdown (when configured)
  • Tokens metrics
  • Performance metrics at both the message and conversation levels
• Prompt Store Configuration: Per-agent prompt store configuration with:
  • Builder integration via Git for version control
  • External database connectivity for better scalability
  • Enhanced versioning and prompt management capabilities
• User Usage Dashboard: Aggregated usage table per user for each AI agent including:
  • Individual user consumption metrics
  • Usage patterns and trends
• Advanced KPI Analytics: Comprehensive analytics system featuring:
  • Dozens of KPIs available per individual agent
  • Admin-level KPIs across all agents
  • Deep adoption analysis without requiring custom dashboards
  • Built-in analytics for comprehensive performance monitoring
• Enhanced Source Display: Improved visualization and presentation of information sources

​

Studio Enhancements

• SSO Integration: Profile edition is now automatically disabled when authenticated with SSO
• User-Space Request: Fixed fetching and caching issues

​

Runtime & Improvements

• Socket Support: Enhanced socketId support within child Broker for better real-time communication
• Memory Optimization: avoid memory leaks in broker package
• Vertical Scaling: enhance NodeJS multithreading support with RUNNER_MAX_THREADS environment variable
• Context Synchronization: Resolved run context synchronization issues that occasionally broke AI Store response processing

​

AI Store & Conversations

• Database Optimization:
  • Renamed collections for better clarity (“wonderful-treefrog-80” → “Conversations”, “brave-zebra-50” → “Messages”)
  • Added multiple indexes to Conversations & Messages collections
• Message Processing:
  • Fixed message sources displaying and broken markdown format
  • Simplified message chunk processing and removed duplicate calls
  • Improved initialization by removing init block chat event

​

AI Knowledge & AI Insight

• Improvements:
  • Disabled validateArguments on many internal automations
  • Removed live message updates listener to lighten processing workload
  • Enhanced saveVectorizeFailed errors with specific public error messages
• API Key Management: Users can now regenerate API keys directly from the UI
• Elasticsearch Support: AI Knowledge is now compatible with Elasticsearch Vector Store

​

AI Builder

• New Components:
  • Block Icon now available in Page Builder
  • New Collapse block for better content organization
  • KPI block enhancements
  • New DateRangePicker block
• Enhanced Functionality:
  • Schema Form fields can use ui:options to add blocks as suffix into Input
  • DropDown blocks can display blocks list as content and be closed with events
  • Datatable columns now accept blocks attribute for full customization
• Code Editor: Fixed slow performance issues in Builder code editor

​

Developer Tools & Runtime

• New DSUL Helpers: Added unsafejson, URL, sanitize, uuid, and power math (^) helpers
• Fetch Improvements:
  • Added allowSelfSignedCert parameter for fetch instruction
  • Enhance Support for HTTP(S)_PROXY variables
  • Automatic authentication on /apps endpoints
• Monitoring: New runtime prometheus metrics for interactions, automations, users, and performance

​

Multi-Platform Support

• Audio & Image: Fixed Vertex AI audio & image support for some models
• Mock Models: Added mock-embeddings and mock completions models for testing