Update requirements and improve model loading logic

switched to qwen

Enhance model loading and configuration. Added BORDERLESS_PRELOAD_MODEL to .env.example and updated README.md for clarity. Refactored device resolution in model.py for improved GPU handling and logging. Added warning filter for UserWarning in ui/__init__.py.

Refactor ZeroGPU handling and update documentation. Reduced default BORDERLESS_GPU_DURATION to 60 seconds in .env.example and README.md. Enhanced error messaging in app.js and gradio_api.js for better user feedback on quota issues. Updated GPU_DURATION logic in config.py to enforce minimum and maximum limits.

Updated the chat UI and backend so MiniCPM thinking renders

MiniCPM5 ZeroGPU migration

Fix string formatting in tool_display_title function for URL handling

fix empty research findings

Compact Chat Tool UI is implemented

Todo schema updated, Carousel UI scroll fix

Richer Globe Research Popups is implemented

parallel research carousel UI

Formatted Chat And Map Details

Structured Final Synthesis

Robust Planner Discovery Workflow

nodes modularized

langgraph integration

todo: langgraph integration

Surface think/reasoning as Thinking messages and add tool args/results to metadata.log in tools.py and respond.py

Make /chat a generator: stream_chat in server_api.py, generator api_chat in app.py, chunked final-answer yields in streaming.py

Point auth prompts at the Server-mode login link.

Refactor response handling and tool execution in the agent module for improved clarity and maintainability.

The UI folder restructured