🚀 AnythingLLM v1.9.1 is live! Update now →
Desktop Changelogs
v1.9.1

Notable Improvements: 🚀

Windows Installer Optimization
Fixed bug where installation time on Windows would take a very long time to complete. Installation time is now significantly faster.

MCP Support Improvements
Refactored MCP support. Patches issues with MCPs not starting or not working correctly.

Chat Input Persistence
Chat input now persistent when navigating between workspaces if not sent when navigating away.

Realtime YouTube Scraping Realtime scraping of YouTube videos is now supported.

Ask @agent to scrape a YouTube video and it will be used to answer your question.

Other Improvements

  • Internal Ollama version bumped to 0.13.0 (opens in a new tab)
  • Managed NVIDIA NIM has been removed from AnythingLLM Desktop.
  • Dell Pro AI Studio model URL updated to new specification.
  • Improved error handling for MCPs not starting or not working correctly.
  • General language improvements and fixes.
  • Keyboard navigation of slash command when showing slash command list.
  • Paperless NGX data connector support
  • Agent workspace system prompt can now use system variables for variable expansion.
  • Use eval_duration from Ollama for accurate TPS calculations.
  • Add SerpAPI web search as agent web-search provider
  • Support AWS Bedrock API key connection method
  • ZAI LLM provider support
  • Anthropic prompt caching and config
  • Ability to set global default prompt for new workspaces
  • Add base64 document attachment support for chat API
  • SSL bypass for local confluence
  • GiteeAI LLM provider support
  • OpenRouter Embedder support
  • Ollama batch embedding support

Bug Fixes

  • Fixed runtime issue with Ollama and LMStudio model caching causing model list to be empty or incorrect.
  • Fixed bug where the MCP panel was not scrollable for certain models.
  • Fix relevance score not showing for Astra, QDrant, Zilliz, and Weaviate citations
  • EPub upload for certain file layouts were failing on upload - this is now fixed.
  • Fixed bug where Gemini thinking output was not showing in chat or hanging the response.
  • Fix infinite loop logic in GitLabLoader
  • Add Svelte renderer to markdown output
  • Disable Prisma CLI telemetry
  • Patch Ollama broken thought output from chat template update
  • Extend HTTP TTL on extension requests for timeout
  • Fix undefined result in llm-instruction blocks
  • Fix directOutput causing hanging response for agent flow calls when streaming
  • Fixed Chroma cloud limitations on payload size for upsert of embeddings.

Pinned Download Links

Revision 1.9.1:

Overviewv1.9.0