
What's New:

- Agent Flows for easier agent skill creation (more blocks coming to desktop soon)
- Built in native OCR for scannedPDF and Images!
Improvements:
- QNN NPU model engine bumped for faster loading of models and inference.
- Include
reasoning_content
in Generic OpenAI connector - Include reasonsing for Deepseek API
- Changed onboarding flow to be native language for system
- Tokenizer performance improvements for large documents
-
<thinking>
Inherit UI font size from UI for think - Azure O1,O3 support and reasoning
- Enable
num_ctx
in ollama embdder to match dimensions - Patch PPLX timeouts + inchat citations from PPLX
- Improved agent logging for web scraping
- Patch Gitlab sub-project pulling
- PAT on Confluence connector
- In chat citations for Perplexity API
Bug Fixes:
- Fixed bad upload UI loop on document loader
- Fixed light mode SQL connector
X
being invisible - Fixed bad codeblock header size on Windows
- Fixed O3 model temperature being present in requests resulting in 403