connectai/PATCHNOTES.md

Patch Notes - v2.2.46

🚀 Key Improvements & Bug Fixes

1. AI Communication Protocol Fix

• Issue: Previously, when a file was attached, the text prompt was completely overwritten by the visionContent structure, leading to empty messages being sent to LM Studio.
• Fix: Merged the text prompt with the vision content array so that both text and file metadata are correctly transmitted.

2. Autonomous Loop Optimization

• Issue: Broad keyword matching (e.g., "조사", "설명") triggered local file analysis tasks even for general questions, causing the agent to bypass the LLM and give incomplete answers.
• Fix: Refined the isProjectAnalysisRequest patterns to be more conservative, ensuring generic conversational requests are always handled by the AI model.

3. Second Brain Profile Management

• Issue: Adding new brain profiles was inconsistent because a virtual "default-brain" (injected in memory) was being saved into the permanent settings, causing profile list corruption.
• Fix:
  • Decoupled runtime virtual profiles from persistence logic.
  • Implemented direct settings access for profile addition to avoid stale config caches.
  • Fixed UI sync issues immediately after adding a new brain folder.

4. Build & Reliability

• Removed premature empty stream chunks that were causing UI flickering.
• Verified build stability with v2.2.46 VSIX package.

---

v2.8.0 (2026-04-29) - Intelligent Collaborative System Evolution

🤖 Multi-Agent Workflow (MAW)

• Planner Agent: 전략 수립 및 작업 단계 자동 설계.
• Researcher Agent: 지식 베이스(Second Brain) 심층 분석 및 데이터 정제.
• Writer Agent: 최종 보고서 형태의 고품질 답변 생성 및 통합.
• 복잡한 요청(분석, 보고서 등) 감지 시 자동으로 멀티 에이전트 모드 활성화.

🔮 Proactive Suggestion Engine

• Behavioral Tracking: 사용자 UI 체류 시간(Dwell Time) 기반 의도 감지.
• Smart Tips: 설정, 지식 동기화, 에이전트 선택 등 상황에 맞는 능동적 도움말 제공.

🎨 UX & Reliability Improvements

• Visual Feedback: 가이드라인에 따른 상태별 아이콘 및 피드백 일관성 강화.
• Model Fix: 모델 선택 및 저장 로직 안정화 완료.
• Agent Handoff: 각 에이전트 간 데이터 전달 및 진행 상황 시각화 개선.

---

Date: 2026-04-25 Version: 2.2.46