Agent Hub Release notes¶

Version 1.3.1 - Apr 20th 2026 ¶

New feature: Agent Chat export recipe¶

An export recipe has been added to Agent Chat, similar to the Agent hub export recipe.

My Agents¶

The new ‘My Agents’ projects, will embed documents using a window mode with 5 pages and an overlap of 1 to improve the performances.

Chat interface¶

Fixed missing title generation issues: Title generation now begins at the start of the streaming process, leveraging only the user’s query.

Miscellaneous¶

Fixed ‘NotRequired’ import path for Python < 3.11 compatibility.

Version 1.3.0 - Apr 14th 2026 ¶

New feature: Agent Chat¶

Agent Chat is a dedicated web app to chat directly with either a model or an agent, offering a quick-to-set-up chat interface.

New feature: Chat to dataset in My agents¶

Agents in the ‘My Agents’ section can now be connected to SQL datasets, making it easier for end users to explore and interact with trusted data they already have access to through workspaces and data collections.

Prerequisites:

This feature requires the Dataset SQL Query tool, available in Semantic Models Lab plugin version 0.2.0 or later.
Only SQL datasets can be connected to agents, and they must belong to a workspace and/or a data collection
To activate this capability, webapp administrators must enable Enable datasets in Agent Hub admin settings and allow browsing from workspaces and/or data collections.

New feature: Localization¶

Agent Hub is now available in Japanese and French, helping organizations support a broader range of users and improve adoption across international teams. Users can select their preferred language from the left panel in Preferences.

New feature: Support multimodal outputs from agents¶

Images and other multimodal outputs generated by agents are now displayed directly in the chat interface and can be downloaded, improving the usability of multimodal workflows and making visual outputs easier to review and share.

New feature: Agent context configuration settings¶

A new advanced settings section in the Admin panel gives administrators more control over the context shared with agents, making it easier to balance richer agent behavior with governance and security requirements.

Bring Your Own Document¶

New feature: Agents now support attachments, making it easier for users to bring their own content into conversations and work with richer business context.

Human-in-the-loop input edition¶

New feature: When ‘Allow editing’ is enabled for a DSS tool, users can directly edit tool call arguments in the conversation, giving them more control, reducing friction, and making human-in-the-loop workflows more efficient.

Left panel¶

The left sidebar has been redesigned to improve navigation and make key actions more accessible. Enhancements include:
- A refreshed sidebar design for a clearer, more intuitive experience
- Replacement of the Favorite section with Quick Access, giving users faster access to favorite agents and the 3 most recently used agents
- Direct access to language settings from the sidebar
- A virtual tour to help users quickly adapt to the new experience

Chat interface¶

Code block rendering improvement: In the chat interface, refreshed the code blocks styling and fixed over flow issue
Added capability to preview file attachments preview image inside the conversation
Fixed single agent issue preventing other agents loading, improving application reliability and reducing the impact of isolated failures.
Fixed infinite loop when users opened a conversation on a slow connection and switched to another one before the first had finished loading. This improves stability and delivers a smoother user experience in low-bandwidth or high-latency conditions.
Fixed bar charts generation issues.

Chart generation¶

Added capability for webapp administrators to provide additional instructions to the LLM during chart generation, making it easier to tailor outputs to business-specific needs and improve how charts align with organizational use cases and reporting expectations.

Logging¶

Improved error logging and set application logging level to DEBUG by default.

Version 1.2.7 - Mar 26th 2026 ¶

Requires DSS >= 14.4.0

Chat interface¶

Fixed document upload error: missing conv_id or user_id

Version 1.2.6 - Mar 13th 2026 ¶

Requires DSS >= 14.4.0

Miscellaneous¶

Fixed CVE alert by updating langchain version.

Version 1.2.5 - Mar 6th 2026 ¶

Requires DSS >= 14.4.0

Chat interface¶

Fixed conversation title generation on very small LLM models.

Miscellaneous¶

Addressed dependency security alerts.

Version 1.2.4 - Feb 27th 2026 ¶

Requires DSS >= 14.4.0

New feature: Human-in-the-loop support for structured agents¶

Human-in-the-loop workflows are now supported for structured agents.

Admin panel¶

Enterprise agent settings have been improved with drag-and-drop reordering.

Chat interface¶

Fixed an issue that prevented PNG, JPG, and JPEG images from being uploaded in conversations.
Fixed an issue where the wrong augmented LLM could be triggered in certain cases.

Settings¶

Table prefixes in the web app settings now support uppercase letters and are automatically converted to lowercase.

Version 1.2.3 - Feb 20th 2026 ¶

Requires DSS >= 14.4.0

New feature: Container-based execution¶

Container-based execution is now supported.

Startup experience¶

Improved handling when the application fails to start by showing a clearer modal message and automatically redirecting administrators to the Admin settings.

Guardrails¶

Document guardrails now also evaluate document titles against defined text and REGEX rules, in addition to extracted body content.

Published agents¶

Fixed an issue where deleting a document from a published agent did not properly remove it from the published knowledge base, ensuring documents stay synchronized after re-publishing.

Miscellaneous¶

Improved error handling in draft mode by surfacing backend errors directly in the UI instead of leaving the interface hanging.
Fixed an issue where the fallback title “New chat” was not persisted when the LLM failed to generate a conversation title.

Version 1.2.2 - Feb 13th 2026 ¶

Requires DSS >= 14.4.0

New feature: Download structured agent artifacts¶

Artifacts generated by structured agents are now downloadable, either directly from the chat interface or through the dedicated download panel.

Export recipe¶

Fixed an empty date field in the export recipe caused by a type mismatch.

Miscellaneous¶

Improved the design of the “Look & feel customization” tab in the Admin settings.
Improved the tool call validation card design.
Adjusted markdown rendering for conversation messages.
Improved conversation creation so user input is not blocked by title generation, especially when using a reasoning model.

Version 1.2.1 - Feb 9th 2026 ¶

Requires DSS >= 14.4.0

New feature: Short-term memory¶

Agents now support short-term memory through memory fragments when enabled at the agent level.

Chat interface¶

Fixed missing action (including copy and thumbs up/down) buttons below assistant responses.

Version 1.2.0 - Feb 9th 2026 ¶

Requires DSS >= 14.4.0

This release includes a schema upgrade. In the event of an upgrade issue, our support team can help resolve it.

New feature: Human-in-the-loop support¶

Conversations with agents now include validation steps for tool actions that require explicit user confirmation, giving you more control and clarity.

New feature: Remote PostgreSQL support¶

Added support for remote PostgreSQL databases, enabling centralized storage of conversations, Admin settings, and related data.

New feature: Reasoning model visibility¶

Improved support for reasoning models by displaying real-time reasoning “chunks” while the agent or LLM is processing a query.

New feature: Document content guardrails¶

Administrators can now enable guardrails within “Core Settings” to enforce content policies using forbidden words or regex. Documents that violate these policies are blocked from reaching the LLM provider.

For attachments, forbidden documents are automatically filtered out of conversations without being deleted from the system.
For Quick Agents, documents are screened during creation and automatically deleted if they violate the configured policies. During conversations, agents containing forbidden documents are filtered out.

New feature: Third-party agent and tool indicators¶

Third-party agents and tools, such as Snowflake and Bedrock, are now more clearly identified in the interface.

New feature: Prompt Library integration¶

The Prompt Library now sources content directly from collections stored in the Enterprise Asset Library of the DSS instance.

New feature: Interface customization¶

Global interface customization is now available for all web app users, including:

White labeling: Set a default primary color, homepage title, and logos.
Asset management: Choose a managed folder for storing assets.

Admin panel¶

The Enterprise Agents tab in Admin settings has been redesigned for better usability.
Real-time Presence badges have been added to indicate when multiple administrators are viewing the same page.
Selection components have been standardized, and search inputs have been added to improve the user experience.

Agent context¶

More information about the caller is now added to the agent context, including dkuOnBehalfOfEmailAddress and dkuOnBehalfOfDisplayName.

My Agents Sharing¶

Fixed an issue where group restrictions were applied even when the setting was disabled. Agents can now be shared with any user in the DSS instance when restrictions are turned off.

Traces explorer¶

Fixed a bug that caused data to go missing in the Traces Explorer.

Draft conversations¶

Document uploads are now disabled in draft conversations to prevent system errors.

Miscellaneous¶

Updated various software packages to address known security vulnerabilities.

Version 1.1.8 - Feb 20th 2026 ¶

Compatibility¶

Fixed compatibility issues between Flask and Flask-SocketIO by restricting the application to supported versions, preventing unexpected errors.

Version 1.1.7 - Feb 19th 2026 ¶

SQL tool¶

Fixed an issue where empty responses from the Agent SQL tool could cause conversations to stop unexpectedly.

Version 1.1.6 - Feb 9th 2026 ¶

Chat interface¶

Restored conversation message actions after renaming a conversation and navigating between conversations. This fixes an issue where copy, thumbs up, thumbs down, and see details buttons could disappear.

Version 1.1.5 - Feb 9th 2026 ¶

Visibility¶

Fixed an issue in agent visibility filtering that could allow users without permission to see unintended agents. Access control is now correctly enforced.

Version 1.1.4 - Feb 5th 2026 ¶

Version 1.1.3 - Jan 16th 2026 ¶

My Agents Sharing¶

Fixed an issue where users could only share agents with users belonging to their DSS groups. By default, agents can now be shared with anyone in the instance, and group restriction is now opt-in (through admin settings).

Security¶

Agent Hub now includes dkuCallerTicket in the agent context, enabling tools to support end-user impersonation when required. This works only with DSS >= 14.3.2.

Version 1.1.2 - Feb 9th 2026 ¶

Requires DSS >= 14.2.0, or DSS >= 14.3.0 if you want to enable VLM annotation option in BYO

Migrations¶

Fixed an issue where table migrations failed when upgrading from v1.0.4 or earlier to v1.1.1.

Miscellaneous¶

Resolved a security vulnerability in the Langchain core dependency (CVE-2025-68664).

Version 1.1.1 - Dec 19th 2025 ¶

Requires DSS >= 14.2.0, or DSS >= 14.3.0 if you want to enable VLM annotation option in BYO.

Conversation Attachments¶

The conversation attachments panel now displays the extraction mode, supporting both Vision and Text modes.
Webapp administrators can now choose the VLM model used for image attachments in Admin settings.

Draft mode¶

Fixed an issue where the conversation was cleared after switching tabs in draft mode.

Chat interface¶

Fixed a bug where sources were temporarily displayed incorrectly while the agent response was streaming.

Tracking¶

Fixed WT1 tracking issues in standalone mode and improved WT1 event counting for the number of agents.

RAG¶

Fixed an issue where RAG LLMs were called with the wrong ID, causing failures.

Version 1.1.0 - Dec 8th 2025 ¶

Requires DSS >= 14.2.0, or DSS >= 14.3.0 if you want to enable VLM annotation option in Bring Your Own Document (BYO).

This release introduces significant improvements to document handling (BYO), a revamped Admin Panel, enhanced sharing capabilities, and numerous UX updates.

New feature: Bring Your Own Document¶

If enabled by the webapp administrator, users can upload documents directly into the conversation context. (attachment button + drag & drop). Admin settings now include extraction mode configuration and VLM annotation support, which requires DSS >= 14.3. The document experience in conversations has also been improved.

Note: this works only with LLMs, not agents.

New feature: Export recipe¶

An export recipe and custom dataset capability have been added to export the data from the SQLite store into datasets.

Chat interface¶

New feature: Chat directly with the LLM: If enabled by the webapp administrator, users can choose to chat directly with the LLM instead of using agents.
Allow users to disable agents: If enabled by the admin the user can choose to directly chat with the LLM (no agents).
Chat input light redesign, moved switches and information inside the Agents button dropdown
Note: If “Document Upload in Conversation” is enabled, users will be able to chat with the LLM. If you want to completely disable this behavior, you must disable Document Upload in Conversation

Admin panel¶

A new Admin Settings panel is available: Centralized and simplified settings directly within the web application and not the Edit tab anymore.
Settings are updated dynamically: After admin save, the updated settings are pushed to every user and instantly reloaded.

Prompt Library¶

Customization options have been added to the Prompt Library in Admin settings.

My Agents Sharing¶

Agent sharing has been improved with more granular options, allowing agents to be shared with specific groups or users within a user’s groups.
An “Exclude group” option has also been added for “My Agents” sharing.

My Agents¶

My Agents can now use tools that were shared from other projects and made available in the current project (Requires DSS >= v14.3.0).

Knowledge banks¶

Added support for jsonSnippet knowledge bank responses, including multi-column retrieval.

Error handling¶

End users now receive an error message when an agent fails to respond or when an error occurs on the server side.

Chat interface¶

Resolved Japanese input issues.
Fixed prompt input focus loss when creating an agent.
Fixed screenshot display for non-textual documents in Sources.

Logging¶

Fixed an issue where the used_agent_ids database column was left empty.

Miscellaneous¶

Addressed pentesting findings and fixed dependency security alerts.

Version 1.0.4 - Nov 7th 2025 ¶

Chat interface¶

Fixed duplicate message rendering after editing a message. Previously, when a message was edited, it appeared twice—once before the assistant’s response was received. This duplicate is now properly removed.

Error handling¶

Enhanced error messages:
- If the backend fails to start or another error occurs, a dialog now appears stating:
  “The application failed to load. Please make sure the service is running or check the logs for more details.”
  
  If the application loads successfully but access is restricted, an Access Denied message is displayed instead.

LLM prompts¶

Fixed an issue where optional instructions from the global system prompt were not being passed to the LLM.

Version 1.0.3 - Oct 24th 2025 ¶

Miscellaneous¶

The documentation link in the plugin description now correctly points to the latest Agent Hub guide.

Version 1.0.2 - Oct 20th 2025 ¶

Augmented LLM¶

Fixed a regression where Augmented LLM settings were no longer being applied.

Version 1.0.1 - Oct 17th 2025 ¶

Permissions¶

Fixed a crash that occurred when a user tried to access a project they didn’t have permission for (which previously crashed the thread retrieving agents).

Settings¶

Renamed the tab referenced in the warning message when a mandatory “User Agents” setting is missing.
Missing mandatory settings no longer block the save action.

Miscellaneous¶

Enabled “copy/open trace” functionality for all profiles.

Version 1.0.0 - Oct 14th 2025 ¶

Requires DSS >= 14.2.0

New feature: Enterprise Agents¶

Users can seamlessly chat with any agent available within the DSS environment, based on your web app setup. Query examples can also be defined to help users get started more effectively.

New feature: My Agents¶

Once enabled by the webapp administrator, users can create, edit, and share their own agents. Once in ‘My Agents’:

They can use Prompt library: A comprehensive library to quickly kickstart the creation of your agents.
Smart mode can automatically prefill certain fields based on the agent instructions.

New feature: Agents Library¶

The Agents Library provides a complete view of available agents, including enterprise, personal, and shared agents, directly inside the web app.

New feature: Charts generation¶

Charts can be generated automatically or on demand based on agent response artifacts.

New feature: Monitoring¶

A Monitoring dashboard allows you to track agent usage effectively through the Monitoring Dashboard available in the Agent Hub.

New feature: Dataiku Stories¶

Users can craft compelling Dataiku Stories directly from your interactions with agents.

New feature: Feedback¶

Users can provide feedback on agent responses.

New feature: Agents as tool & Agents in parallel¶

Agent Hub can query multiple agents simultaneously, either by using agents as tools or by executing parallel calls, depending on the configuration.

New feature: Security tokens¶

Agent Hub sends user information (DSS groups, login, email) as part of the context to agents, safeguarding retrieved data.

New feature: Sources & activity¶

Users can view the sources used by agents and inspect agent activity to better understand generated responses.
Artifacts generated by agents can also be viewed and downloaded.

New feature: Traces¶

Users can copy traces or open the Traces Explorer directly for deeper analysis.

New feature: Mobile responsive design¶

The interface is fully responsive, providing a seamless experience across desktop and mobile devices.

Agent Hub Release notes¶

New feature: Agent Chat export recipe¶

My Agents¶

Chat interface¶

Miscellaneous¶

New feature: Agent Chat¶

New feature: Chat to dataset in My agents¶

New feature: Localization¶

New feature: Support multimodal outputs from agents¶

New feature: Agent context configuration settings¶

Bring Your Own Document¶

Human-in-the-loop input edition¶

Left panel¶

Chat interface¶

Chart generation¶

Logging¶

Chat interface¶

Miscellaneous¶

My Agents Sharing¶

Chat interface¶

Miscellaneous¶

New feature: Human-in-the-loop support for structured agents¶

Admin panel¶

Chat interface¶

Settings¶

New feature: Container-based execution¶

Startup experience¶

Guardrails¶

Published agents¶

Miscellaneous¶

New feature: Download structured agent artifacts¶

Export recipe¶

Miscellaneous¶

New feature: Short-term memory¶

Chat interface¶

New feature: Human-in-the-loop support¶

New feature: Remote PostgreSQL support¶

New feature: Reasoning model visibility¶

New feature: Document content guardrails¶

New feature: Third-party agent and tool indicators¶

New feature: Prompt Library integration¶

New feature: Interface customization¶

Admin panel¶

Agent context¶

My Agents Sharing¶

Traces explorer¶

Draft conversations¶

Miscellaneous¶

Compatibility¶

SQL tool¶

Chat interface¶

Visibility¶

Navigation¶

My Agents Sharing¶

Security¶

Migrations¶

Miscellaneous¶

Conversation Attachments¶

Draft mode¶

Chat interface¶

Tracking¶

RAG¶

New feature: Bring Your Own Document¶

New feature: Export recipe¶

Chat interface¶

Admin panel¶

Prompt Library¶

My Agents Sharing¶

My Agents¶

Knowledge banks¶

Error handling¶

Chat interface¶

Logging¶

Miscellaneous¶

Chat interface¶

Error handling¶

LLM prompts¶

Miscellaneous¶

Augmented LLM¶

Permissions¶