Download Links
Configuration Changes & Integration Updates
- Ollama Version: Bumped to v0.13.0. Uses
eval_duration for TPS calculations. Supports batch embedding.
- NVIDIA NIM: Managed NIM has been removed.
- Dell Pro AI Studio: Model URL updated to new specification.
- Agent Web Search: Added SerpAPI provider.
- AWS Bedrock: Added API key connection method.
- ZAI: Added LLM provider support.
- GiteeAI: Added LLM provider support.
- OpenRouter: Added Embedder support.
- Anthropic: Added prompt caching support.
- Telemetry: Prisma CLI telemetry disabled.
Features & Connectors
- Data Connectors: Added Paperless NGX support.
- Chat API: Added base64 document attachment support.
- Local Confluence: Added SSL bypass toggle.
- YouTube Scraper: Supports real-time scraping via
@agent.
- Workspace System Prompts: Supports system variables expansion.
- Global Prompts: Option to set default prompt for new workspaces.
- UI/UX: Keyboard navigation for slash commands; chat input persistence during workspace navigation.
Bug Fixes & Patches
- Windows Installer: Optimized installation time.
- MCP: Refactored MCP support to fix startup issues; fixed non-scrollable MCP panel for certain models.
- Model Caching: Fixed empty/incorrect lists for Ollama and LMStudio.
- Gemini: Fixed hanging responses and missing thinking output.
- Ollama: Patched broken thought output from chat template update.
- Relevance Scores: Fixed visibility for Astra, QDrant, Zilliz, and Weaviate citations.
- EPub: Fixed upload failures for specific layouts.
- GitLabLoader: Fixed infinite loop logic.
- Chroma: Patched payload size limits on embedding upserts.
- Markdown: Added Svelte renderer.
- Timeout: Extended HTTP TTL on extension requests.
- Agent Flow: Fixed hanging response caused by
directOutput during streaming.
- LLM Instructions: Fixed undefined result in instruction blocks.