v7.0.0 - Agent Edition: Hybrid Automation & Research
Describe this release
This release marks a major architectural shift from a standard AI wrapper to a modular Autonomous Agent. ZAI v7.0 introduces hybrid workflows, allowing the execution of tasks that require both terminal commands and GUI interactions.
🚀 Major Features
- Hybrid GUI Automation: Implemented
GUIAutomationBridge. ZAI can now control mouse/keyboard and interact with desktop applications when terminal commands aren't enough (usespyautogui+ Vision). - Web Research Engine: Integrated
duckduckgo-search. ZAI can now perform live web searches to find documentation, solve errors, or fetch real-time data. - Visual Intelligence: Added
ImageAnalyzerusing Gemini Vision. Users can pass image paths for analysis (e.g., "analyze error.png"). - P2P Terminal Sharing: Native TCP socket implementation with optional Ngrok support for remote pair programming/assistance.
- Modular Architecture: The system now utilizes a "Graceful Degradation" approach. If optional dependencies (like
torchorpyautogui) are missing, the shell continues to function in Core Mode without crashing.
🔧 Technical Improvements & Tweaks
-
Temperature Tuning:
-
Online Lightning Mode: Reverted temperature to
0.0to ensure maximum determinism and speed for quick commands. -
Offline Mode: Local Phi-2 model temperature adjusted to
0.1for better stability in local inference. -
Intent Detection: Added a dedicated AI router to intelligently switch between Terminal, GUI, Research, or Hybrid modes based on user prompts.
-
Self-Healing Strategy: Enhanced error recovery logic that attempts up to 5 different strategies (shell switching, encoding fixes) before failing.
📂 Housekeeping
- Legacy v6.0 core has been moved to the
legacy/directory for backward compatibility.