Skip to main content

​
July 2026

​
Release 07.24.26 — July 24, 2026

​
Chat & Personas

  • Thread-local chat attachments: attachments are scoped strictly to the thread they are uploaded in and are never written to the Knowledge Graph or shared Knowledge Base, keeping them isolated from other users and other threads
  • Attachment sources in the response panel: attached files selected by the planner are now surfaced alongside Knowledge Base citations in the Sources panel and persisted to response metadata
  • Attachment filenames in conversation history: planner-selected filenames are appended to the stored query record, enabling accurate semantic retrieval over past turns that involved file attachments (streamed responses are unaffected)
  • Per-file token budget for large attachment sets: replaced the previous single-clip strategy with a two-pass allocation that reserves per-file header overhead and distributes the remaining budget evenly, so every attached file is represented in context regardless of attachment count
  • Web search sources in the Sources panel: web results the model actually used are now surfaced alongside Knowledge Base and attachment sources for full provenance on web-grounded answers
  • Chunk ID suppression in responses: response assembly now prohibits internal chunk identifiers from appearing in user-facing output, preventing raw UUIDs from leaking into the streamed response

​
Integrations

  • Redshift schema expansion: removed view-type filtering from Redshift schema introspection so the full schema surface — including views and additional schema objects — is available to Enterprise users with Redshift connectors

​
Datasets

  • Truthful OCR failure propagation: extractions where all pages fail OCR now return a typed error instead of an empty document with inflated quality scores; partial extractions report quality proportional to successful pages, and failure markers no longer pollute embeddings
  • VLM image normalization: high-resolution images are now normalized to safe dimension and pixel bounds before submission to the Vision Language Model, preventing request timeouts across the OCR extractor and the synchronous and asynchronous processors
  • Per-attempt VLM retry deadlines: each VLM retry attempt now receives an independent timeout with capped, interruptible backoff, replacing a shared deadline that caused all retries to fail instantly after the first timeout

​
Thread Deletion

  • Response-engine data purge: deleting a conversation thread now triggers a cascading, asynchronous purge across PostgreSQL (conversation record, messages, summaries, attachments), Qdrant (per-tenant conversation embeddings scoped to the deleted thread and its owner), and the hot in-memory cache
  • The internal deletion route is idempotent and ownership-enforced: unknown or already-deleted threads are handled gracefully, and foreign-owned data cannot be removed

​
Observability

  • WebSocket query metrics: query count and duration are now recorded on the WebSocket path, closing a gap where normal chat traffic produced no signal in Grafana
  • Per-stage span instrumentation: full per-stage tracing across orchestrator, embedding, context-assembler, and model-selector stages, with correct lifecycle handling for success, fallback, stream interruption, client cancellation, and disconnect (interrupted streams are excluded from the completed-query counter)
  • OpenTelemetry off-by-default and crash-proof: telemetry providers only initialize when an OTLP endpoint is explicitly configured, initialization failures roll back to a disabled state instead of crashing the service, and exports run in background threads with a bounded timeout so they never block the request path. The Helm chart OTEL block is conditionally rendered for deployments that omit OTEL configuration entirely

​
Under the hood

  • Centralized model ID configuration: the active model identifier is now an environment-configurable knob with semantic role constants for planning, synthesis, and system default, replacing hardcoded string literals across all call sites. The active model can be changed by updating a single environment variable and restarting the service
  • Summary embeddings in orchestrator planning: top document summaries ranked by vector similarity against the current query are now retrieved from the per-tenant summary collection at query time and injected into the orchestrator planning prompt, improving the relevance and precision of the retrieval plan. Includes stability fixes in the summarization-engine storage and API layer

​
Release 07.22.26 — July 22, 2026

​
Integrations

  • Added Azure Database for PostgreSQL: connect using OAuth or your database credentials
  • Redshift: schema filtering improvements for data connections

​
Chat & Personas

  • Web search sources: sources used in web-search answers now appear in the Sources panel alongside your documents, with each source shown as a clickable link reflecting only what the answer actually used
  • Chat attachments: URLs and integrations are now supported as attachment sources, plus minor fixes and improvements

​
Under the hood

  • UI improvements and bug fixes

​
Release 07.15.26 — July 15, 2026

​
User Interface Updates

  • Refreshed look and feel across the platform: updated fonts and typography, refined themes, and an improved mobile experience
  • Deprecated Docs mode in chat; a link to docs.4minds.ai has been added to the profile icon dropdown

​
Chat & Personas

  • Added chat attachments: attach files directly in chat threads (first release — the feature will continue to evolve)

​
Deployment

  • AWS Marketplace: the 4MINDS deployment package is now available via AWS Marketplace, with prerequisites and setup documentation

​
Under the hood

  • API improvements, stability enhancements, performance optimizations, and bug fixes

​
June 2026

​
Integrations

  • Added S3 and Dropbox dataset sync: connect S3 buckets and Dropbox folders directly as synced dataset sources
  • Fixed dataset sync failures (“connection is closed”) on slow-responding integrations like OneDrive and Box
  • Fixed Google Drive sync getting stuck in a loop on certain folder structures
  • Fixed integration file attachments not attaching correctly in Advanced Mode

​
Datasets

  • Added end-to-end support for CSV, XLSX, XLS, and Parquet files in the Wren/DuckDB structured data pipeline, including GPT-OSS model support
  • Added new text and markdown accumulating chunkers for higher-quality document ingestion
  • Increased uncompressed ZIP upload limit to per-tier settings (up to 2,000 MB), replacing the previous 100 MB cap
  • Fixed ZIP upload validation errors around file size and file count
  • Set a 30-minute minimum dataset sync interval to prevent runaway sync loops
  • Capped the graph visualization at 500 files per view to keep large graphs responsive

​
Chat & Personas

  • Fixed custom persona descriptions not reaching the model prompt on the REST path
  • Fixed user messages disappearing immediately after send when starting a new conversation
  • Smaller, collapsed-by-default Sources dropdown for a cleaner chat experience
  • Fixed duplicate files appearing in the Sources list
  • Fixed “Add Data” in the right panel not working in fullscreen mode
  • Fixed dataset share confirmation message being lost after sharing

​
API

  • /v1/chat/completions now correctly handles array-form message content for OpenAI-compatible clients
  • OpenAI integration now checks deleted-model status before running inference

​
Admin

  • Added a toggle to show or hide 4MINDS internal users in the Admin usage view

​
Authentication

  • Fixed an error blocking new user signups

​
Legal

  • Updated Privacy Policy content

​
May 2026

​
User Interface Updates

  • Refreshed login and sign-up pages with updated styles (May 1)

​
SYMI

  • Launched SYMI versioning: save snapshots of your SYMI workspace, view a changelog of what’s changed, and reset the workspace back to a clean state with a confirmation prompt before anything is wiped (May 5)
  • Added PDF renderer and LibreOffice renderer: SYMI can now generate PDF files (May 5)

​
Datasets

  • Added .trs file support: users can now add .trs files to their datasets (May 13)

​
April 2026

​
Integrations

  • Added Google Workspace (GSuite) integration for access to Gmail, Drive, Calendar, and Docs (April 3)
  • Added Gong integration for revenue intelligence and sales conversation data (April 3)
  • Added Amazon RDS integration for relational database connectivity and structured data access (April 3)
  • Added BigQuery integration for cloud data warehouse access and large-scale analytics (April 3)

​
Developer Tools

  • Launched 4MINDS MCP Server (April 15)

​
User Interface Updates

  • Launched SYMI AI Assistant with a major interface and control center overhaul (April 17)

​
March 2026

​
Teams

  • Launched Teams Tab: organization administrators can now create and manage teams, assign users, and organize access to shared resources based on team membership
  • Added Shared Resources View: users can view models and datasets shared with them by other organization members, accessible directly from the Teams tab
  • Added Granular Resource Sharing: models and datasets can now be shared with the entire organization, specific teams, or individual users

​
Role-Based Access Control (RBAC)

  • Launched Multi-Layer Role Hierarchy: roles now operate across three levels — organization (Owner, Admin, Billing Manager, Member, Guest), team (Owner, Admin, Member, Viewer, Guest), and system - with permissions cascading accordingly
  • Added Access Levels: users are assigned Basic, Plus, or Admin access tiers based on their subscription plan, controlling feature availability across the platform
  • Added Custom Roles & Granular Permissions: admins can grant or revoke individual resource:action permissions for any role, scoped at the organization, team, or resource level
  • Added Resource-Level Access Control: per-model and per-dataset access grants can be targeted at specific users, teams, or the entire organization — with support for time windows, token limits, rate limits, and usage duration restrictions on model grants
  • Added RBAC Audit Trail: all role and permission changes are logged with full context including actor, target, resource, IP address, and timestamp
  • Added SSO Auto-Provisioning: SSO groups can now be mapped to teams with a default access level, enabling automatic role assignment on login across Azure AD, Okta, Google, GitHub, and LDAP

​
February 2026

​
Agentic Platform

  • Launched Agentic AI Platform: rotating agent status indicators with staggered timing for real-time visibility into agent activity
  • Added Extended Thinking Display: “Extending Thinking…” indicator during response regeneration
  • Improved Sources Experience: sources now displayed in a dropdown with pagination for cleaner chat interface

​
External Model Integrations

  • Added Amazon SageMaker Integration: create models on the 4MINDS platform directly from your existing SageMaker models
  • Added Amazon Bedrock Integration: connect to AWS-managed foundation model endpoints hosted on Amazon Bedrock
  • Added Microsoft Foundry Integration: create models from external models deployed on Microsoft Foundry

​
Integrations

  • Added Databricks Serverless Updates: OAuth/M2M authentication and Personal Access Token (PAT) support for secure, flexible connectivity

​
January 2026

​
User Interface Updates

  • Launched Major UI Refresh with significant visual and structural improvements across the platform (January 20)
  • Added AI Assistant Panel: dedicated left sidebar with contextual onboarding and help system for guided workflows
  • Introduced Enhanced Model Info Panel: redesigned right sidebar consolidating Quick Actions, model details, and easy access to evaluations, API keys, and model settings
  • Added DOC MODE Toggle: new interface mode for enhanced document-focused workflows
  • Implemented Three-Column Layout: improved organization and workflow efficiency across the platform
  • Enhanced Visual Canvas: updated graph rendering with force simulation and viewport culling for better performance
  • Streamlined Evaluations View: updated evaluations table with improved status indicators and action menus

​
Integrations

  • Added Supabase integration for database and backend access (January 21)
  • Added ServiceNow integration for IT service management workflows (January 21)
  • Added Slack integration for team communication and messaging data (January 21)
  • Added Google Drive integration for cloud document access (January 21)
  • Added Dropbox integration for file storage and sharing (January 21)
  • Added Splunk integration for log and security data analysis (January 21)

​
Authentication

  • Added Single Sign-On (SSO) with AWS (January 21)
  • Added Single Sign-On (SSO) with Ping Identity (January 21)
  • Added Single Sign-On (SSO) with Okta (January 21)

​
December 2025

​
Integrations

  • Added Salesforce integration for CRM access to leads, accounts, contacts, opportunities, and sales analytics (December 9)
  • Added HubSpot integration for CRM data, contact management, conversations, and email tracking (December 4)

​
November 2025

​
New Features

  • Added Log Timeline Chart for visual monitoring: view system activity over time with a color-coded timeline displaying log event distribution, severity levels (Debug, Info, Warning, Error, Critical), and activity patterns to quickly identify spikes or gaps before diving into detailed logs (November 25)
  • Added Model as Judge evaluation method: automatically compare your customized model against a base foundation model with ChatGPT acting as an AI judge to evaluate responses side-by-side, providing detailed analysis of factual grounding, key differences, and winner rationale to help identify knowledge gaps and guide training data improvements (November 24)
  • Introduced System Logging for real-time visibility into AI model operations with live log streams, severity-level filtering (DEBUG, INFO, WARNING, ERROR, CRITICAL), search functionality, and log management tools including pause/resume, clear, and download capabilities (November 14)
  • Implemented Automatic Data Synchronization (Rsync): datasets created from integrations now automatically sync on login. The platform tracks a manifest of your source files and fetches any new additions, keeping your AI knowledge base current without manual re-imports (November 13)

​
Integrations

  • Added Microsoft Fabric integration: browse and import from Lakehouses (files) and Warehouses (tables) for unified analytics platform access (November 25)
  • Added Box integration for cloud-stored documents and file management (November 13)
  • Added NetApp integration for enterprise storage access (November 12)
  • Added Office 365 integration for seamless access to emails, calendars, OneDrive files, Teams messages, and OneNote (November 11)

​
Authentication

  • Added Single Sign-On (SSO) with Microsoft (November 11)

​
October 2025

​
New Features

  • Introduced Evaluations functionality to automatically generate comprehensive model performance reports with key metrics, enabling data-driven decisions on model quality (October 31)

​
Development Environment Enhancements

  • Added Cloud Shell support for browser-based command-line access to 4MINDS (October 31)

​
Integrations

  • Added support for 5 new integrations: Amazon S3, CoreWeave Storage, SharePoint, Snowflake, Google Cloud Storage (October 28)
  • Added dataset and file import capabilities from Databricks, Hugging Face, and Azure Blob Storage (October 17)
  • Added credential management for connected integrations (October 16)

​
Web Search & Scraping

  • Added web scraping feature supporting multiple URL inputs for real-time data extraction (October 15)
  • Introduced web search toggle during chat and inference sessions (October 7)

​
File Format Support

  • Added support for 10 new file types : BMP, DOCX, GIF, JPEG, JPG, MD, PDF, PNG, TIFF, XLSX (October 28)
  • Added support for graph exports: SVG, PNG, JSON, GraphML, GEXF, PDF (October 1)

​
Coming Soon

  • Platform evaluation support, providing insights into usability, features, and performance.
  • OCR support for image uploads and dataset integration
  • API key generation for custom application integrations