Deprecations

  • NVIDIA NIM: Phased out in v1.9.0; will be removed in the subsequent release.

Feature Changes & Configurations

Agent System

  • Streaming: Tool calls and agent responses stream in real-time across all providers and models.
  • Web Ingestion: Agent can download and ingest files (PDF, XLS, CSV) directly from web URLs in real-time.

Microsoft Foundry Local Integration (Beta Preview: Windows & macOS)

  • Auto-starts Microsoft Foundry Local on AnythingLLM Desktop launch (if installed).
  • Auto-unloads models to free system resources.
  • Note: Model downloads must be managed via foundry cli; the UI only lists locally downloaded models.

Linux-Specific Updates

  • Ollama Integration: Ships with Ollama (0.11.4) pre-installed for local LLM support.
  • AppArmor (Ubuntu): Auto-creates AppArmor rule for AppImage execution without extra config.
  • Desktop Entry (GNOME): Auto-generates .desktop file for launcher integration.

Core Configurations & Developer API

  • System Prompt Variables: Added variables workspace.name and workspace.id.
  • API Response Source Reporting: Sources are reported in API responses on the final chunk of a stream.
  • Folder GET Endpoint: Patched HTTP response code for developer API.
  • Generic OpenAI Embedder: Added option to configure an artificial delay.
  • Generic OpenAI Requests: Added User-Agent header.
  • Ollama/LMStudio: Automatic detection of model context window size.
  • Ollama Timeout: Added custom HTTP response timeout configuration.
  • PGVector: Added metadata sanitization to prevent bad bytes in jsonb fields.
  • Export formats: Support for exporting JSON/JSONL with attachments from Workspace Chats viewer.