Commit Graph

6 Commits

  • feat: Implement Canvas Isolation Architecture for surgical code precision
    - Replaced old refinement protocol with technical Canvas Isolation Architecture
    - Integrated layered canvas logic (Base, Change, Merge, Validation)
    - Enforced isolation chamber and quarantine zone constraints in prompts
    - Updated QA self-check gate to strictly monitor for scope creep
  • fix: Enable Ollama models for code generation
    - Added getActiveModel() function to automationService for dynamic model selection
    - Replaced all hardcoded 'qwen-coder-plus' strings with getActiveModel() calls
    - Added localStorage sync when model is changed in orchestrator
    - This enables Ollama and other models to work in all automation tasks
  • fix: Ollama click handler and stream completion
    - Fixed sidebar Ollama Cloud section to always open modal on click
    - Improved stream handling with proper buffer flush on end
    - Added 120s timeout for Ollama requests
    - Added better logging for debugging
    - Fixed activeRequest cleanup on errors
  • fix: Fix Ollama API key storage - keytar ESM import and file fallback
    - Fixed keytar ESM import to use .default property
    - Added comprehensive error handling and logging for vault operations
    - Fixed fallback to file-based storage when keytar is unavailable
    - Added debug logs to help troubleshoot key storage issues
  • feat: Add Ollama Cloud integration with 20+ free AI models
    - Added AI Model Manager to sidebar for quick model switching
    - Integrated Ollama Cloud API with official models from ollama.com
    - Added AISettingsModal with searchable model catalog
    - Models include: GPT-OSS 120B, DeepSeek V3.2, Gemini 3 Pro, Qwen3 Coder, etc.
    - Added 'Get Key' button linking to ollama.com/settings/keys
    - Updated README with Ollama Cloud documentation and free API key instructions
    - Fixed ChatPanel export issue
    - Added Brain icon for reasoning models