XendiAI-Plugin/CHANGELOG.md

Change Log

All notable changes to the "Xendi AI" extension will be documented in this file.

[1.2.1] - 2025-10-03

🧹 Packaging Improvements

Bug Fixes

• Fixed "nul" file causing unsafe extraction error
• Removed temporary build folders from VSIX package
• Cleaned up .vscodeignore to exclude development files

Package Optimization

• Removed temp_extract/ and temp_latest/ directories
• Excluded *.log files
• Excluded IMPROVEMENTS_ROADMAP.md (development doc)
• Excluded tsconfig.json.disabled
• Package size reduced from 2.8 MB to 2.73 MB

---

[1.2.0] - 2025-10-03

🌐 Multi-Provider Support - Major Release

✨ 8 AI Provider Support

• LM Studio - Run models locally on your machine (existing)
• Ollama - Simple local AI with easy model management (existing)
• OpenAI - GPT-4, GPT-4 Turbo, GPT-3.5 Turbo (NEW)
• OpenRouter - Access to 100+ models including GPT-4, Claude, Gemini (NEW)
• Anthropic - Claude 3.5 Sonnet, Claude 3 Opus (NEW)
• Cohere - Command R+, Command R (NEW)
• Groq - Ultra-fast inference with Llama, Mixtral (NEW)
• Mistral - Mistral Large, Mistral Medium (NEW)

⚡ Productivity Features

• Real-time Streaming: Streaming responses for all cloud providers (except Ollama)
• Keyboard Shortcuts: 15+ shortcuts for maximum productivity
  • Ctrl+Enter - Send message
  • Ctrl+K - New chat
  • Ctrl+L - Clear chat
  • Ctrl+/ - Focus input
  • Ctrl+Shift+C - Copy last response
  • Escape - Cancel request
  • Ctrl+↑/↓ - Navigate message history
  • Ctrl+1/2/3/4 - Switch tabs
  • Ctrl+Shift+M - Toggle Chat/Agent mode
• Token Counter: Live token counting with cost estimation per model
  • Per-model pricing for GPT-4, Claude, Cohere, etc.
  • Session statistics tracking
  • Accurate estimates for code vs text
• Theme Sync: Automatic light/dark theme sync with VS Code

🎨 UI/UX Improvements

• Provider Selection: AI Provider dropdown in Settings tab
• API Key Management: Secure API key input for cloud providers
• Fixed Minimum Sizes: Consistent UI sizing (600px body, 400px tabs, 200px chat)
• Better Tab Switching: Fixed tab navigation bugs
• Null-safe JavaScript: Comprehensive null checks for all DOM elements

🔧 Technical Improvements

• Provider-specific Authentication:
  • Bearer tokens for OpenAI, OpenRouter, Groq, Mistral
  • x-api-key header for Anthropic
  • Authorization header for Cohere
• Smart Default URLs: Each provider has pre-configured default endpoints
• Default Models: Pre-configured models for Anthropic and Cohere
• Streaming Implementation: Full SSE (Server-Sent Events) parsing for real-time responses
• Optimized Logging: Structured logging for better debugging

🐛 Bug Fixes

• Fixed tab switching not working (ID mismatch resolved)
• Fixed null reference errors in addEventListener calls
• Fixed TypeScript compilation overwriting extension.js
• Fixed streaming cursor animation
• Fixed settings not persisting correctly

📚 Documentation

• Completely overhauled README.md with 8 provider setup guides
• Added keyboard shortcuts documentation
• Added token counter and cost estimation docs
• Updated troubleshooting section
• Provider comparison table

🔄 Migration Notes

• Existing LM Studio and Ollama configurations will continue to work
• New users can choose any of the 8 supported providers
• API keys are stored securely in VS Code globalState
• No breaking changes to existing functionality

---

[1.1.3] - 2025-10-02

🐍 Python Virtual Environment Support

• Automatic venv Detection: Agent now detects venv/, .venv/, and env/ directories
• Auto-Activation: Python virtual environments are automatically activated before running commands
• Seamless pip/python: No need to manually activate venv - just use pip install or python directly
• Smart Command Wrapping: Commands are wrapped with source venv/bin/activate && automatically

🔧 Developer Experience

• Agent knows about venv: Updated system prompt to inform agent about automatic venv activation
• Better Python Development: Agent can now install packages and run Python code without environment issues
• Cross-Project Support: Works with any Python project structure (venv, .venv, or env)

---

[1.1.2] - 2025-10-02

🎯 Agent Improvements

• Fixed Topic Switching: Agent now stays focused on one task until completion
• No More Task Jumping: Agent won't suggest or start additional work unless requested
• Laser Focus: Clear instructions to work ONLY on the current user request

🎨 UI Improvements

• Command Result Dropdowns: All agent tool outputs now display in collapsible dropdowns
• Tool Icons: Visual icons for each command type (🔍 glob, 📄 readFile, ⚡ runShellCommand, etc.)
• Cleaner Agent Chat: Command results are collapsed by default for better readability
• Consistent Design: Dropdown style matches thinking process display

🐛 Bug Fixes

• Fixed agent wandering off-topic during multi-step tasks
• Improved command output formatting in agent mode

---

[1.1.1] - 2025-10-02

🚀 Performance Improvements

• Optimized API Requests: Added connection keep-alive and request timeouts (60s) for faster communication
• Reduced Token Limits: Decreased max_tokens from 2000 to 1000 for faster response times
• Better Server Communication: Improved fetch configuration with abort controllers

🎨 UI Improvements

• Redesigned Statistics Panel: New clean, professional design matching extension theme
• Removed Emojis: Statistics now use text labels (FILES, LINES, TOKENS) instead of emoji icons
• Enhanced Styling: Monospace font for values, uppercase labels, purple accent colors
• Better Layout: Vertical stat items with clear separation and improved readability

🐛 Bug Fixes

• Fixed statistics panel visual consistency with extension design
• Improved stats visibility toggle with CSS classes

---

[1.1.0] - 2025-10-02

✨ New Features

🧠 Thinking Process Display

• AI Reasoning Visualization: Models that output thinking process (using <think> tags) now show expandable dropdown
• Transparent AI Decisions: See exactly how the AI reasons through problems
• Collapsible Interface: Keeps chat clean while making reasoning available on-demand

📊 Agent Statistics & Monitoring

• Real-time Stats Display: See files created/modified, lines added/deleted, tokens used
• Task Plan View: Visual overview of agent's execution plan and progress
• File Diff Links: Clickable file paths with diff view (like GitHub Copilot)
• Performance Metrics: Track token usage and operation efficiency

🔄 Ollama Support

• Multiple LLM Backends: Now supports both LM Studio and Ollama
• Server Type Selection: Easy switching between LM Studio (localhost:1234) and Ollama (localhost:11434)
• Auto-detection: Automatically handles different API formats

💬 Enhanced Conversation Management

• Auto-generated Titles: Titles automatically created from first message
• Timestamp Display: Shows relative time ("5 Min.", "2 days") or full date/time
• Context Preservation: Full conversation history maintained across all messages
• Delete with Confirmation: Safe conversation deletion with German/English prompts
• Quick Actions: "New Conversation" button in chat for instant access

🎨 Markdown Rendering

• Full Markdown Support: Renders code blocks, lists, headers, bold, italic
• Syntax Highlighting: Language-specific code block rendering
• Clean Formatting: Professional display of AI responses

🌐 Context Reset Commands

• vergiss alles (German)
• alles vergessen (German)
• reset context (English)
• clear context (English)
• forget everything (English)

🐛 Bug Fixes

• Fixed conversation context not persisting between messages
• Fixed agent losing context during multi-step tasks
• Improved error handling for network issues

🔧 Technical Improvements

• Better API compatibility (LM Studio + Ollama)
• Optimized conversation history handling
• Enhanced markdown rendering engine
• Improved statistics tracking

---

[1.0.0] - 2025-10-02

🎉 Initial Release

Xendi AI - Autonomous AI Coding Assistant for VS Code

✨ Features

🤖 Agent Mode

• Autonomous task completion - Works like GitHub Copilot/Claude
• Multi-step workflow - Explores, analyzes, fixes, and tests automatically
• Smart tool execution - Automatically reads files, finds bugs, implements fixes
• Continuous iteration - Keeps working until task is 100% complete
• Workspace awareness - Knows project structure and context

💬 Chat Mode

• Interactive AI assistant for coding questions
• Code explanations and suggestions
• Real-time code generation

🛠️ Powerful Tools

• File Operations: Read, write, search, replace
• Code Analysis: Pattern search, bug detection
• Shell Commands: Run tests, builds, install packages
• Web Integration: Google search, web content fetching
• Memory: Save and recall information

🌍 Universal Language Support

• JavaScript/TypeScript
• Python
• Java, C++, C#
• Go, Rust
• PHP, Ruby, Swift, Kotlin
• And many more...

🔒 Privacy & Security

• 100% local execution with LM Studio
• No cloud services
• No telemetry or tracking
• Your code stays on your machine

🎨 User Experience

• Clean, intuitive interface
• Multilingual UI (English/German)
• Comprehensive logging system
• Auto-approve tools for seamless workflow
• Model selection and switching

🔧 Technical Details

• LM Studio Integration: Seamless connection to local LLM server
• Workspace Tools: Full VSCode workspace API integration
• Conversation History: Persistent chat and agent conversations
• Error Handling: Robust error detection and recovery
• Performance: Optimized file operations with caching

📝 Configuration

• Customizable LM Studio server URL
• Language selection (English/German)
• Auto-approve settings for tools
• Model temperature and token settings

🎯 Best For

• Debugging: Find and fix bugs automatically
• Refactoring: Improve code structure and quality
• Feature Development: Build new features end-to-end
• Learning: Understand code and get explanations
• Testing: Run and fix test failures

🙏 Credits

• Built with LM Studio API
• Inspired by GitHub Copilot and Claude
• Community feedback and testing

---

Developer: Robin Oliver Lucas (RL-Dev) License: MIT Website: https://rl-dev.de

---

For support and feedback:

• GitHub Issues: https://github.com/RL-Dev/xendi-ai/issues
• Website: https://rl-dev.de