LazyBug Copilot - Visual Studio AI Coding Assistant Extension
📖 中文版
Product Overview
LazyBug Copilot is a "Cursor-like" intelligent coding assistant extension designed specifically for Visual Studio. It integrates Large Language Model (LLM) capabilities to provide developers with intelligent code creation, refactoring, and Q&A experiences. The extension supports multiple mainstream AI service providers, enabling developers to enjoy AI-assisted programming within their familiar IDE environment.
Version 0.20.1 Release Notes
- Fixed an issue where the GLM model would sometimes interrupt unexpectedly
Version 0.20 Release Notes
- Added copy/paste buttons in the Provider & API settings page
- Added symbol search support for more languages (
*.html, *.css, *.js, *.java, *.py, *.ts)
See patchnotes.md for full version history.
Core Features
- Multi-Turn Intelligent Chat — Markdown-based chat content. Session history stored per VS solution.
- Session Management — Switch between historical sessions, one-click rollback to any previous state, session favorites management, and cost statistics per session.
- Symbol Link Recognition — Clickable symbol links in the chat window for quick navigation to definitions.
- Smart Code Editing — AI directly modifies project files with multi-file support, before/after Diff View, modification tracking, undo/redo , and file backup.
- Automatic Code Database — Automatically builds a code database from all files in your solution with incremental updates.
- Codebase Search — Fast text search for ultra-large projects (million-line scale). Significantly faster than ripgrep, especially in large codebases.
- Symbol Search — Fast symbol search for C/C++/C#/JavaScript/Java/Python/TypeScript codes. Works out of the box — no LSP configuration required.
- Smart Input Box — Tag-based file attachment system with
@ auto-completion, input history (PageUp/PageDown), quick model switching.
- Image Attachment — Paste images directly into the chat input to send to vision-capable LLMs.
- Multi-Model Support — Customizable API endpoints. Supports mainstream LLMs: OpenAI, Anthropic, Google Gemini, OpenRouter, Moonshot (Kimi), z.ai (GLM), DeepSeek and more. Also supports local LLMs (Ollama, LM Studio).
- Multi-API Format — Supports three API formats: OpenAI-compatible, Anthropic, and Gemini.
- Skill System — Browse, create, rename, and toggle skills via a management panel. Supports BuiltIn, Global, and Project-level skills. Allow using AI to edit or create new skills.
- Custom Prompts —
global_rules.txt and project_rules.txt for customized prompts;
- CLI Tool Integration — Execute cmd.exe, bash.exe, python.exe scripts directly from the chat, extending capabilities beyond coding.
- Context Usage Control — Real-time context usage display with 5 context levels. Allow keeping context under relatively low level (< 30k tokens) even in extremely long conversations while maintaining high response quality. Automatic compression/decompression when context level changes.
- MCP Support — Model Context Protocol server support via stdio and URL, with a built-in UI to manage them.
- UI Scaling — Hold
Ctrl and scroll the mouse wheel to freely zoom the chat interface and text size.
Usage Scenarios
- Code Review and Refactoring: Let AI analyze your code and suggest refactoring improvements.
- New Feature Development: Describe your requirements, and AI will assist in generating code frameworks and implementations.
- Bug Fixing: Describe the issue symptoms, and AI will help locate and fix the bug.
- Code Explanation: Ask about complex code logic, and AI will provide detailed explanations.
Usage Tips
Open the Chat Panel: Open the chat panel via the Visual Studio menu View → Other Windows → LazyBug Chat. It is recommended to bind a shortcut to the View.LazyBugChat command for quick access.
Database and Disk Space: The LazyBug chat database is independent of the project and is stored centrally on the C drive. Please ensure sufficient free space on your C drive (for ultra-large projects, more than 10 GB may be required).
Adding File Attachments:
- Select a file in the Solution Explorer, right-click, and choose Add to LazyBug Chat to attach it to the current conversation.
- You can also do the same by right-clicking the file's tab above the code editor.
Symbol Database Construction (C/C++): When you open a C/C++ project for the first time, LazyBug will automatically build the symbol database in the background. For ultra-large projects (e.g., 3 million lines of code), this process may take a significant amount of time (approximately 30 minutes to 1 hour). Symbol query results may not be fully accurate until the build is complete.
Code Comparison (Diff View):
- Click the title of a file editing label in the chat panel to display the Diff view in the main editor; press
Space to hide it.
- Clicking the title repeatedly allows you to quickly jump between different diff hunks.
Avoid Editing Conflicts: Please do not manually edit files while AI is working, especially when it is modifying file contents.
UI Scaling: When the mouse focus is inside the LazyBug chat window, hold Ctrl and scroll the mouse wheel to freely zoom the interface and text size.
Cost Statistics: The cost for each chat turn is calculated based on the unit price entered in the LLM api setting. When the LLM provider uses a complex billing model (such as a subscription plan), the statistics are approximate and for reference only.
Task Breakdown Strategy: LazyBug is not designed to handle extremely large and complex tasks in a single pass. Break your development tasks into smaller, clearly defined sub-tasks for the best AI-assisted experience.
Code Database Updating: After adding a new file to the solution, save the solution file so that the new file is included in the code database.
Skill Usage:
- There are 3 types of skills:
- BuiltIn: Verified skills bundled with the extension. These should generally not be modified.
- Global: Skills shared across all projects.
- Project: Skills specific to the currently opened project.
- Activating a skill does not mean its full content is loaded into the context. To force-load a skill, copy its path and paste it into the file attachment list.
- Many existing skills exist in the ecosystem and may have compatibility issues with LazyBug. You may need to tweak them until they work smoothly.
- You are always encouraged to use AI to edit existing skills or create new ones.
- Install the necessary environments (Node.js, Python, GIT, etc.) to support various CLI commands.
Compression Model Selection: Use the Evaluate button to assess the speed, reliability, and compression quality of the currently selected compression model, helping you choose the most suitable one.
How Context Level Works:
Context Level Settings Consideration:
- Higher context levels consume more tokens and incur higher costs, especially on expensive models.
- However, higher context levels reduce the frequency of compression, which improves cache hit rates and can also lower costs to some extent.
- Currently, context is not compressed within a single Q&A turn to avoid cache miss.
- Compression is not unlimited — it will always preserve a baseline answer quality. So even if you set a low context level, sometimes context usage may still significantly exceed the upper limit.
- That said, compression will more or less degrade the model's answer quality.
- Overall recommendation: For expensive models (Anthropic, GPT), set the lowest context level (Level 1, < 30k tokens), unless you notice a significant drop in answer quality.
Report an Issue
If you encounter any bugs or have suggestions for improvement, please report here. Your feedback is the driving force behind our continuous improvement!
| |
