Skip to main content

Return to docs

What's new in h2oGPTe v1.6.x

· 18 min read

We are excited to announce the release of h2oGPTe 1.6! Read on to learn about the new features and improvements in h2oGPTe 1.6.x series that help you find answers and generate content based on your private data.

Overview of the 1.6.x Release​

The h2oGPTe 1.6.x series delivers significant capabilities for autonomous AI assistance, enhanced APIs, and improved document processing:

Major features​

  • Agent functionality: Autonomous tool use with code execution, placing #1 on the GAIA leaderboard for General AI assistants
  • Role-Based Access Control (RBAC): Comprehensive permission system for users and groups
  • REST API: OpenAPI-compliant interface with auto-generated bindings for Python, JavaScript, and Go
  • Custom GPT creation: Collection-based configuration system for tailored AI assistants

Key improvements & fixes across releases​

  • Updated supported LLMs list: Added coverage of new large language models, including OpenAI's GPT-5 series (v1.6.37)
  • OpenAI compatible API: Added OpenAI API compatibility for integration with existing applications, supporting chat completions, responses, and streaming (v1.6.37)
  • Exception handling and resilience: Improved exception handling and increased resilience for document ingestion and old parsing issues (v1.6.35)
  • Internationalization: Added RTL language support to improve UI compatibility with right-to-left languages (v1.6.34)
  • Enhanced mobile experience: Better navigation, responsive design, and improved card layouts (v1.6.33)
  • Developer experience: Async API examples, simplified client connections, and enhanced code execution (v1.6.33)
  • System reliability: Better error handling, improved self-tests, and enhanced stability (v1.6.32, v1.6.33)
  • Performance optimizations: Faster document ingestion, improved chat queries, and streamlined processing (v1.6.32)
  • Client stability: Improved real-time chat stability and connection reliability for web applications (v1.6.40)
  • Multilingual support: Added translation framework supporting both left-to-right and right-to-left languages and full Spanish language support (v1.6.41)
  • Agent capabilities: Added support for selecting user personas when using an agent and a specialized Data Science agent type for advanced analytics tasks (v1.6.41)
  • Enterprise security: Secret Manager and Secure Connectors now configurable by administrators for centralized credential management and secure external integrations (v1.6.43)
  • Workspaces: Added workspace functionality with tagging and personal workspace migration (v1.6.43)
  • Enhanced agents: New Tool Builder Agent and improved agent response UI (v1.6.43)
  • New connectors: Added new connector to import content directly from Atlassian Confluence (v1.6.43)
  • Improved document management: Added video scene descriptions and auto-tagging with agents (v1.6.43)

Major patch releases

h2oGPTe v1.6.43​

Improvements​

Workspaces and organization​

  • Workspace functionality: Added workspace feature to organize and manage your content with tagging and tracking capabilities.
    • Existing personal content automatically migrates to the new workspace system

Agent capabilities and tools​

  • Tool Builder Agent: New specialized agent type designed for building and customizing tools
  • Automatic model routing: Agents now automatically select the best model based on accuracy requirements for optimal results
  • Agent monitoring: Added tracking for agent conversation turns to help monitor agent usage
  • Final Answer Mode: New dropdown in agent configuration to control how agents deliver final responses
  • Agent streaming and files UI: Improved real-time streaming interface and file management for agent responses
  • Human-in-the-loop planning: Added control to review and approve agent planning before execution
  • Document auto-tagging: Agents can now automatically tag documents using specialized tools
  • Tool defaults: Option for admins to enable tools by default when creating new agents for faster setup
  • Authentication improvements: Improved authentication tab interface for configuring agent credentials

New model support​

  • Google Vertex AI Gemini 2.5 Pro: Added support for Google's latest Gemini model via Vertex AI
  • GPT-5 Codex: Added support for OpenAI's GPT-5 Codex model

Document processing and connectors​

  • Video scene descriptions: h2oGPTe now generates automatic scene descriptions for better understanding of video content
  • Auto-Sync Collection Connector: New connector that automatically synchronizes collections with external sources
  • SharePoint OAuth2 authentication: Enhanced SharePoint connector with OAuth2 user authentication for improved security

User interface and experience​

  • Clickable table links: Tables now support clickable links for better navigation
  • Microphone animation: Added visual animation when microphone is active during voice input
  • Chat session creator: Users can now see who created each chat session on chats and collection pages
  • Shareable extractors: Extractors page can now be shared with other users
  • Collection application: Improved interface for applying collection settings in chat sidebar
  • Chat with specific collection: Added ability to start a chat directly with a specific collection
  • Document deletion: Users can now delete documents from collections when they have proper permissions

Privacy and security​

  • Custom PII detection: Added AI-powered detection for user-defined personal information labels

User experience refinements​

  • Global settings discoverability: The global settings are now under System Settings tab for easier access
  • Action menus on inactive chats: Added action menu buttons for archived or inactive recent chats
  • Searchable user dropdowns: User selection dropdowns now support search for easier user finding
  • Searchable model options: Model selection dropdowns now support search functionality
  • Dark mode improvements: Fixed font color issues in dark mode for better readability
  • Android voice improvements: Enhanced speech synthesis for Android webview users
  • Role selector: Added role selector to the side menu in the user interface.

Interface improvements​

  • Disabled input when recording: Text input automatically disabled when microphone is active to prevent conflicts
  • Title truncation: Long titles are now automatically truncated with ellipsis for cleaner display

Performance improvements​

  • Faster collection loading: Reduced API calls for collections and documents, resulting in faster page loads
  • Better document retrieval: Markdown and HTML content now processed more effectively for improved search and retrieval

Bug fixes​

User interface​

  • Layout fixes: Fixed padding and spacing issues in primary and secondary panels
  • Code viewer: Removed width restrictions on code viewer for better code display
  • File status badges: Fixed positioning of file status badges
  • Height issues: Resolved height calculation problems in various interface components
  • Markdown tables: Improved handling of edge cases in markdown table rendering
  • UI crash prevention: Fixed crash that occurred when no language models were configured

h2oGPTe v1.6.0

Released: January 31, 2025

Agent features​

Agent overview​

The h2oGPTe Agent enables autonomous tool use through code execution. It uses large language models (LLMs) for code generation and reasoning. The agent achieved #1 ranking on the GAIA leaderboard, which measures General AI assistant usefulness.

Key features include:

  • Deep Research assistance: Provides autonomous analysis with full transparency into decision-making processes
  • Comprehensive output: Delivers analysis summaries, internal chat transcripts, and downloadable artifacts for each conversation
  • File management: Gives you access to newly created files (PDF, Excel, PowerPoint) and all code snippets used in document creation

Agent control​

You can enable or disable the agent through the chat input interface. The agent operates autonomously based on your prompts and configured prompt templates.

Accuracy Presets control conversation depth and processing time:

  • Quick
  • Basic (default)
  • Standard
  • Maximum

Each preset defines the number of conversation turns and maximum processing time per turn.

Agent tools​

Administrators can enable or disable these agent tools for chat queries:

Code and Development:

  • Aider Code Generation
  • Shell Scripting
  • Python Coding

Data Visualization:

  • Mermaid Chart-Diagram Renderer
  • Image Generation

Content Processing:

  • Ask Question About Image
  • Audio-Video Transcription
  • Convert Document to Text
  • Screenshot Webpage

Research and Search:

  • Google Search
  • Bing Search
  • Scholar Papers Search
  • Wolfram Alpha Math Science Search
  • Wikipedia Articles Search
  • Wayback Machine Search
  • Web Image Search

Document Analysis:

  • Ask Question About Documents
  • RAG (Retrieval-Augmented Generation) Vision
  • RAG Text

System Integration:

  • H2O Driverless AI Data Science
  • Browser Navigation
  • Download Web Video
  • Advanced Reasoning
  • Evaluate Answer

Network Access:

  • Internet Access
  • Intranet Access

Access control and permissions​

Role-based access control (RBAC)​

This version introduces a comprehensive role and permission system. Each role contains specific permissions, and administrators can assign roles to users and groups from federated authentication providers like LDAP.

Available permissions:

Chat management:

  • Delete chats
  • Submit chat feedback

Collection management:

  • Add collections
  • Delete collections
  • Edit collections
  • Make collection public
  • Share collections

Document management:

  • Add documents
  • Delete documents

Template management:

  • Delete prompt templates
  • Edit prompt templates
  • Share prompt templates

System administration:

  • Show admin center
  • Allow device pairing when configured
  • Show extractors
  • Show live logs
  • Show models page
  • Show private button
  • Manage roles
  • Display system notifications
  • Display developer settings

API and developer tools​

REST API​

A new REST API complements the existing Python RPC client. The API conforms to the OpenAPI standard and provides built-in Swagger UI documentation.

Auto-Generated bindings:

  • Python REST API
  • JavaScript REST API
  • Go REST API

Custom GPT creation​

Create custom AI assistants using the formula: Collection + Collection Settings + Default Chat Settings = Custom GPT.

Each collection contains default chat settings that apply to new conversations. You can apply current settings as collection defaults through the Apply current settings as collection defaults button or via API.

Code generation for chat messages​

Each chat message displays the equivalent Python client code, enabling developers to replicate queries programmatically.

Model and processing improvements​

Reasoning model support​

The models page displays reasoning capabilities and shows which reasoning models support non-reasoning models, similar to vision model relationships. Reasoning models work with chat, RAG (Retrieval-Augmented Generation), and agent use cases.

Vision capabilities​

Enhanced vision model functionality across the platform.

Document processing​

Parsing improvements:

  • Layout detection
  • Chunking algorithms
  • Image captioning
  • Text conversion
  • Document highlighting
  • Excel handling (large tables are summarized while remaining fully accessible to agents for data analysis)

Handwriting recognition​

The H2O Mississippi model provides default handwriting-to-text transcription and ships with the platform.

Supported LLMs​

Support for proprietary and open-source models includes:

Cloud providers:

  • Claude 3.5 (Bedrock)
  • OpenAI o1 (Azure)
  • OpenAI o1-mini (Azure)
  • Gemini 2.0 Flash
  • Gemini 2.0 Flash Thinking

Open-Source Models:

  • DeepSeek V3
  • DeepSeek R1
  • MiniMaxAI
  • Qwen/Qwen2.5
  • Qwen/Qwen2-VL
  • Qwen/QwQ
  • Llama-3.3-70B
  • Llama-3.2-11B-Vision
  • Llama-3.2-90B-Vision

H2O Models:

  • H2O Mississippi

Performance and scalability​

Architecture improvements​

  • Models Service: Redesigned backend enables horizontal scaling for document ingestion and chat through a dedicated service shared by chat, crawl, and core services
  • Auto-Scaling: Optional KEDA-based auto-scaling for the models service
  • Database Operations: Parallelized Vex database operations
  • Conversion Speed: Accelerated text-to-PDF conversion

User experience enhancements​

Interface improvements:

  • Separate guardrails and PII (Personally Identifiable Information) settings
  • GUI-based custom guardrails configuration
  • Enhanced PDF display
  • Collection thumbnails
  • Improved models page layout
  • Better scrolling and pagination
  • Syntax highlighting for markdown and code blocks
  • Enhanced job cancellation
  • Faster automatic RAG type detection

Model testing​

Self-Test enhancements:

  • Functional self-tests with multimodal RAG and guided JSON
  • Better detection of model endpoint configuration issues

Administrative features​

White labeling​

Customization options include:

  • Custom logos
  • Color schemes
  • Greeting messages
  • Personality configuration in prompt templates

Topic modeling​

Generate topic model visualizations for any collection with a single click. Visualizations show clusters of similar phrases and concepts, providing content overviews and identifying areas for content optimization.

Collection management​

Lifecycle controls:

  • Configurable collection expiration times
  • Collection size limits

Performance optimizations​

  • Faster public chat sharing

Security​

No critical or high CVEs at the time of release.

Support and resources​

For technical support and questions about this release:

Additional resources​

Upgrade information​

We recommend upgrading to the latest version of h2oGPTe 1.6.x to access these improvements. The upgrade process preserves all your existing data and configurations.

Feedback