We are excited to announce the release of h2oGPTe 1.7! Read on to learn about the new features and improvements in the h2oGPTe 1.7.x series that expand agent control and introduce enterprise-grade governance capabilities.

Overview of the 1.7.x Release​

The h2oGPTe 1.7.x series expands agent control with Human-in-the-Loop feedback and persistent Memory Blocks, improves RAG performance, and introduces enterprise-grade API governance and fairness controls alongside significant UI enhancements.

Major features​

• Human-in-the-Loop (HITL) & Agent Control: Active model guidance and feedback mechanisms combined with comprehensive behavior customization options for AI agents.
• Memory Blocks: Persistent context blocks that agents read from and write to across sessions, with group-based access control and automatic sharing when collections are shared.
• Native MCP Integration: Standardized Model Context Protocol server for connections to external tools, chat, and document collections, with sub-tool filtering, an inspection UI, and a standalone tool marketplace page.
• Enterprise Governance & Fairness: Per-user API rate limiting, automated API key deactivation triggers, and fairness mechanisms to prevent system resource monopolization.

Key improvements & fixes across releases​

• Performance & Speed: Significantly faster document indexing, document ingestion, and parallel file downloads. Real-time RAG progress messaging now counts unique documents rather than raw chunks. (v1.7.0)
• Security & Guardrails: Hardened MCP server security. Enhanced AI guardrails by adding collection metadata to context, with improved UI feedback for guardrail violations. (v1.7.0)
• Scheduled Task Reliability: Multiple race condition and state management fixes for scheduled tasks, including correct pause behavior, timestamp advancement, and manual Run Now on paused tasks. (v1.7.1)
• Auto Memory Blocks: Persistent memory blocks are now automatically created for chats without manual configuration. (v1.7.1)

Major patch releases

h2oGPTe v1.7.0

New features​

Agents and tools​

• Human-in-the-Loop (HITL) course correction: You can now actively guide and provide feedback to the model during generation.
• Advanced Agent controls: Fine-tune agent behaviors with comprehensive new customization options.
• Agent tool creation: Create new tools directly from agent files using an improved UI workflow.
• Memory Blocks: Persistent memory blocks let agents store and retrieve context across sessions. Blocks can be set as the default for new chats, shared with groups, and are automatically shared when their parent collection is shared.
• Native MCP server: A native Model Context Protocol (MCP) server built with the Go SDK provides standardized access to chat, documents, and collections.
• MCP sub-tool filtering and inspection: Filter available MCP sub-tools through SDK options and inspect tool definitions through a dedicated UI.
• MCP tool marketplace: A dedicated Tools page lists all available MCP tools as a browsable marketplace.

RAG and AI models​

• New models added: Support for Claude 4.6 Sonnet*, Claude 4.6 Opus*, Gemini 3.1 Pro Preview, GPT-5.2*, and three NVIDIA Nemotron models via OpenRouter: Nemotron-3-Super-120B-A12B, Nemotron-3-Nano-30B-A3B, and Llama-3.3-Nemotron-Super-49B-v1.5.
• Fast Agentic RAG: A performance-optimized mode for Agentic RAG that reduces retrieval latency.
• RLM RAG: A Recursive Language Model generation approach that iteratively refines retrieval and generation steps for improved answer quality.
• Agentic RAG improvements: Enhanced agentic RAG capabilities for more reliable multi-step retrieval workflows.
• RAG streaming progress: Real-time progress indicators appear during RAG operations.
• RAG inclusion/exclusion filters*: Apply include and exclude filters on chat queries to control which documents are used in retrieval.
• Smart chat history: The model now decides whether a query requires context from the chat history.
• Model tiering system: Organize and manage models by tier for streamlined model selection.
• Model notifications: You are automatically notified when new models are added to the platform.

Chat and collaboration​

• Chat branching: Branch off into new conversations from specific messages within an existing chat session.
• Chat processing visibility: View your exact queue position during chat processing with real-time feedback.
• Fairness throttle notifications: Receive a notification when your session is throttled, with automatic reconnection if the chat connection drops.
• Auto-generated turn titles*: The system automatically generates descriptive titles for chat turns.
• Enhanced citations*: Specific passages from citations are now visually highlighted, and references display as pills for better readability.
• Slow chat notifications: You receive notifications when chat or crawl jobs take longer than expected.
• Quote-to-ask: Highlight any text in a chat response and instantly quote it into the message input to ask a follow-up question in context.

Scheduled tasks​

• Scheduled Tasks UI: A new dedicated Scheduled Tasks page lets you create, view, update, and delete scheduled AI workflows directly from the UI; no API or scripting required.
• Execution tracking: Monitor each task's status, last-run timestamp, and execution history. Tasks can be set to run on a recurring schedule and paused or resumed at any time.
• Automated AI workflows: Combine a prompt, a collection, and a schedule to run repeating AI queries automatically — useful for recurring summaries, monitoring, or report generation.
• /schedule slash command: Start a new scheduled task directly from the chat input using the /schedule slash command.

Documents and collections​

• Collection management: Pin favorite collections for quick access, and access Most Recently Used (MRU) collections in the sidebar.
• Collection import in Agent File Explorer: Import collections directly from the agent file explorer.
• Collection settings restructure: Chat and ingestion settings are now integrated into collection settings for a unified configuration experience.
• Enhanced document indexing: Phase-based job progress tracking for indexing operations.
• Eval Studio integration: Topic Modeling is now integrated directly into Eval Studio.
• Enhanced feedback export: Export feedback by collection and date range.
• Collection evaluations table: Collection evaluations now include clear, descriptive labels.
• Confluence support*: The Confluence connector now fully supports attachments.
• File version history: Documents now track version history. Upload a new version of a file and h2oGPTe groups it with previous versions, letting you browse, compare, and manage the full revision history of any document.

User interface​

• KaTeX toggle: Switch KaTeX math rendering on or off in preferences.
• PDF viewer original mode*: View PDF documents in their original format.
• Inline images in side panel: Open inline images and file links in the side panel without leaving the current view.
• Artifact Viewer full-screen: Full-screen support for the Artifact Viewer.
• Chat drag-and-drop: Improved chat drag-and-drop capability.
• Dynamic dialog height: Dynamic height adjustment for the Add Document dialog.
• Per-form field reset: Reset individual form fields without clearing the entire form.
• PII toggle alert: An information alert displays when toggling PII detection settings.
• System settings tabs and search: The system settings page now supports tabs and search.
• Customizable reference highlight colors: Customize the colors used to highlight referenced passages.
• Extractor UI expansion: Expanded extractor interface with additional fields and custom field support.
• Autosync connector jobs page: A new UI page for monitoring autosync connector jobs.

Authentication and security​

• API key management: Administrators can track detailed API key usage and control permissions.
• API key auto-deactivation: Keys are automatically deactivated when an administrator revokes the associated user's permissions.
• Per-user API rate limiting: Set and enforce per-user rate limits on API requests.
• OAuth improvements: Improved OAuth dropdown rendering.
• Signup abuse prevention: New signup abuse prevention mechanisms with administrator bypass options.
• System notifications: Administrators can create and manage system-wide notifications that are displayed to all users across the platform, useful for announcing maintenance windows, policy changes, or important updates.

Improvements​

Performance and processing​

• Significantly faster indexing: Major performance upgrades to document ingestion and indexing, with phase-based job tracking that shows progress per indexing phase.
• Real-time visibility: View your exact queue position during chat processing and watch real-time RAG streaming progress, now counting unique documents instead of raw chunks.
• Collection lifecycle management: Import collections with lifecycle settings and use the skip_reparse* option to import without re-parsing content.
• Copy linked connectors: Collection owners can copy linked connectors to new collections.

User interface and experience​

• Upgraded file viewing: The PDF viewer now supports an original view mode and displays page numbers. Inline images and file links open in the side panel.
• Artifacts and code: The Artifact Viewer supports full-screen mode, and KaTeX rendering can be toggled on or off.
• Workspace fluidity: Improved chat drag-and-drop capability, dynamic height adjustments for dialogs, and a responsive sidebar aligned to standard breakpoints.
• Sidebar rename: The sidebar Docs link is renamed to "Docs & Guides."

Security and guardrails​

• Improved guardrail violation feedback: Better UI feedback when guardrail violations occur.
• Streaming with prompt guard: Streaming is now maintained when only prompt guard is enabled.

Language and content handling​

• Enhanced character support: Expanded UTF-8 support for CJK, Hebrew, and Hindi characters in text files.
• Markdown handling: Improved handling of Markdown formatting within the PII model, and Markdown files are now previewed natively as text.
• Chromium SVG-to-PDF conversion: Improved SVG-to-PDF conversion using Chromium.

Python client​

• Adds comprehensive job control functions* (pause, stop, play, finish).
• Adds user list pagination and the ability to retrieve public collection permissions.
• Adds support for vector database migration and custom metadata keys.

Bug fixes​

Chat and UI​

• Fixed reversed roles when exporting or downloading chat history.
• Fixed chat table viewport overflow issues.
• Fixed the share button width in full panel mode.
• Fixed single dollar signs ($) being interpreted as math formula delimiters.
• Fixed file auto-switching behavior.
• Fixed the Include chat history setting not being honored when configured at the collection level.
• Fixed label clicks incorrectly triggering guardrail deletion.
• Fixed the OCR model and Audio input language fields being swapped in the upload dialog.
• Fixed the self-reflection toggle being incorrectly disabled in Agent Mode.
• Fixed the 401 insufficient permissions page display.
• Fixed delete message tooltip not appearing.
• Fixed dictated content leaking between messages.

AI and processing​

• Fixed chat history not being used when vision is enabled.
• Fixed guardrail-violating content appearing in chat history.
• Fixed agentic RAG failing in the same chat session when using agent-only files.
• Fixed import jobs not failing when PDF conversion fails.
• Fixed incorrect cost display when no vision model is configured.
• Fixed text-to-speech not working in certain configurations.
• Fixed OCR text insertion into PDF documents.
• Fixed document thumbnails failing to generate for certain image formats.

Connectors and integrations​

• Fixed Confluence connector null OAuth token issue.
• Fixed GCS connector stability issues that caused document ingestion failures.
• Fixed connector settings button visibility based on permissions.

Security and permissions​

• Fixed collection-level permissions for chat deletion.

Support and resources​

For technical support and questions about this release:

• Documentation
• REST API Guide
• Agent tool configuration
• Email: support@h2o.ai

Upgrade information​

We recommend upgrading to the latest version of h2oGPTe 1.7.x to access these improvements. The upgrade process preserves all your existing data and configurations.

---