Skip to main content

September 24 2026 Update

· 2 min read
H&Z digital
H&Z GenAI Platform Team

The September 24 update adds welcome messages for the workspace and for individual agents, rich formatting in prompt templates, and much longer reads of uploaded documents.

Highlights​

  • Add a shared greeting for your workspace, or give each agent its own formatted introduction
  • Use headings, lists and links in prompt templates and agent example prompts
  • Read more of large uploads, with improved handling of document sections and continuation

Welcome Messages​

  • Administrators can set a general chat welcome message; an agent's own message takes precedence
  • Welcome messages support formatted text and an HTML source view; existing plain-text and Markdown greetings are converted to formatted text, and an agent's welcome message can be checked in the agent editor's live preview
  • Added a warning before converting HTML to rich text when formatting could be lost
  • Improved the empty-chat layout on desktop and mobile, and fixed saved or cleared greetings being replaced by older content

Prompts and Templates​

  • Added rich-text editing and display for prompt templates and agent example prompts
  • Improved preservation of prompt content when selecting, copying, searching and editing
  • Fixed missing values in the category edit form, and improved updating prompt templates when a category is renamed, including retries after a partial failure

Agents and Knowledge Bases​

  • Knowledge-base search in Settings matches descriptions again, and the knowledge-base API gained optional paging, search and sorting
  • Stabilized agent ordering and prevented duplicate rows when loading more agents or knowledge bases
  • Fixed agent selections carrying over when switching between knowledge bases

Files and Documents​

  • Raised the default extracted-text limit from 30,000 to 1,000,000 characters
  • Improved how much uploaded text can be included for models with larger context windows
  • Fixed document section reads stopping early or miscounting lines, and improved full-text reads of PDF and Word documents

Chat Settings and Reliability​

  • Output-token settings now follow the selected model's limits without overwriting your saved preferences
  • Improved the visibility of conversation-loading errors
  • Improved usage and cost attribution for API requests and image generation