v0.9.4

Highlights:

• Upstream error messages stored and shown permanently
• Parses and displays cost statistics from /user endpoint
• Added docling processing support for .docx files

v0.9.3

Highlights:

• Added speech generation for received messages
• Improved message editor
• Clarified tool usage, disabled for external models
• Increased modularization for developers
• Bug fixes

v0.9.2

Highlights:

• Added folders and search for organizing and finding conversations
• "Suggest user prompts" option to provide follow-up message suggestions
• New "summarize conversation and replace" button
• Developer can individually enable/disable UI modules, including "tools", "memory", "suggest prompts"
• Improved or fixed several tools, including arcana and web_search
• Improved audio generation model
• Basic support for audio input and transcription via tools using whisper
• Possible to select and copy UI text in multiple areas
• Bug fixes

v0.9.1

Highlights:

• Storing and denoting model name as well as token usage for each message
• Redesigned tools menu with support for additional tools
• Added UI indicator for private vs. non-private mode when using external models and services
• Bug fixes and improvements to localizations and usability

v0.9.0.1

• Bug fixes

v0.9.0

Highlights:

• Redesigned UI with the ability to collapse the left and right sidebars for maximum chatting space. Optimized for smoothness and arbitrarily large conversations and attachments.
• New model selector in the top center with the ability to display more models at once. Switched default model to Qwen 3 30B A3B Instruct 2507 for tool support and performance.
• Audio input with supported models, such as Qwen 2.5 Omni 7B model
• Support for GWDG tools to extend models' capabilities (works well with Qwen 3 30B A3B Instruct 2507)
  • Web search tool with Google
  • Image generation and modification
  • Speech generation
  • Arcana/RAG with any GWDG-local model
  • Support for external MCP Server
  • Realtime updates as tools are being used
• User can now backup all data in a single JSON file with the "Export Data" button in the user settings
• Small improvements, including better code sanitization to prevent cross-site styling, smoother auto-scrolling, new menu to navigate to other AI services, unified attach button, and more

v0.8.1

Highlights:

• New "memory" feature: Chat AI can learn and remember relevant details from all conversations
• Memory is only stored locally in the browser
• Can now type in model selection menu to search for specific models
• Customizable global "timeout" to prevent endless-loop responses
• Added Latex rendering option besides Markdown and Plaintext
• Default settings can now be defined in configuration file
• Fixed rendering bug in last line of text before references
• Generated arcana links now open a new conversation

v0.8.0

Highlights:

• Added selectable Personas in the UI, loaded from https://github.com/gwdg/chat-ai-personas
• "Info" messages supported in imported conversations and personas
• Simplified running and configuring Chat AI as standalone UI, compatible with SAIA API key
• Improved import/export function for conversations

v0.7.4

Highlights:

• Fixed syncing issues when multiple Chat AI tabs are open
• Updated URL parsing for arcanas
• Support rendering references as links

v0.7.3

Highlights:

• Settings panel replaced with hideable sidebar
• Attachments displayed as boxes with thumbnails in prompt
• Added attachment support for more file types
• Version number displayed in footer
• "key" field removed from arcana settings