AnyRAG Pilot

AI-powered semantic search for VS Code with GPU acceleration

AnyRAG Pilot brings enterprise-grade Retrieval-Augmented Generation (RAG) to your development workflow. Index your workspace, GitHub repositories, and any content - then search with natural language.

🎯 Two Ways to Use

@anyrag Chat Participant - Focused RAG Search

Use when you want answers ONLY from your indexed content

• ✅ Guaranteed to search your indexed code/docs only
• ✅ Always shows source attribution with similarity scores
• ✅ No contamination from general LLM knowledge
• 🎯 Perfect for: "How does MY authentication work?" "What's in MY config?"

GitHub Copilot + MCP - Flexible AI Assistant

Use when you want Copilot's full capabilities with optional RAG

• ✅ Copilot decides when to use your indexed content
• ✅ Combines your code with general programming knowledge
• ✅ Broad capabilities beyond just code search
• 🌐 Perfect for: General coding help, best practices, mixed context questions

✨ Features

• 🎯 @anyrag Chat Participant - Targeted search of ONLY your indexed content with source attribution
• 💬 Chat Indexing - Index conversations with /indexchat (Free: 1 chat, Pro: unlimited)
• 🚀 GPU Accelerated - CUDA/MPS support for lightning-fast embeddings
• 🔒 Privacy First - All processing happens locally on your machine
• 📚 Index Anything - Workspaces, GitHub repos, documentation, chat conversations
• 🔌 Model Context Protocol - Integrates with GitHub Copilot as MCP server
• 💾 Persistent Storage - Indices survive across sessions

💎 Pro Features ($10/month)

• ✨ Unlimited indexed documents (Free: 1000 docs)
• 📦 Unlimited indexed sources (Free: 3 sources)
• 💬 Unlimited chat indexing (Free: 1 chat)
• 🎨 Custom embedding models - Use any HuggingFace model (Free: 3 presets)
• 🗂️ Multiple indices - Organize content by project or use case
• 🏷️ Chat naming & management - Rename and organize indexed conversations
• 🎯 Priority support

Upgrade to Pro →

🚀 Quick Start

1. Install the extension from VS Code Marketplace
2. Wait for initial setup (first launch only) - The extension will install Python dependencies, which may take 2-5 minutes. You'll see a progress notification.
3. Index your workspace: Ctrl+Shift+P → AnyRAG Pilot: Index Workspace
4. Ask focused questions: Open chat and use @anyrag how does authentication work in this codebase?
   • @anyrag searches ONLY your indexed content and shows sources
   • Regular Copilot chat can use indexed content OR general knowledge
5. Index conversations: In chat, use /indexchat to save your conversation for future search

💬 Chat Commands

Use these commands in the @anyrag chat participant:

• /indexchat - Index the current conversation with an auto-generated name
• /indexchat my-chat-name - Index with a custom name for easy reference

Managing Indexed Chats:

• View chats: Ctrl+Shift+P → AnyRAG Pilot: Show Indexed Sources
• Rename chats: Select a chat source → Rename Chat
• Re-run /indexchat anytime to update with new messages (replaces old version)

📋 Requirements

• VS Code 1.90.0 or higher
• GitHub Copilot Chat extension (required for @anyrag chat participant)
• Python 3.13+ (auto-detected or configure in settings)
• 4GB+ RAM (8GB+ recommended for large indices)
• Optional: NVIDIA GPU with CUDA for acceleration

🔧 Configuration

Community Tier Settings

• anyragPilot.embeddingModel - Choose from 3 preset models:
  • all-MiniLM-L6-v2 (default) - Fast, 384d
  • all-MiniLM-L12-v2 - Balanced, 384d
  • all-mpnet-base-v2 - Best quality, 768d
• anyragPilot.pythonPath - Manual Python path (auto-detected by default)
• anyragPilot.enableGPU - Enable GPU acceleration (default: true)
• anyragPilot.searchResults - Number of search results (default: 20)

Pro Tier Settings

• anyragPilot.embeddingModel - Select "custom" to use any HuggingFace model
• anyragPilot.customEmbeddingModel - Enter model name (e.g., BAAI/bge-large-en-v1.5)

🔍 Finding Compatible Embedding Models (Pro)

✅ Compatible models must have:

• sentence-transformers library tag on HuggingFace
• Model type: "Sentence Transformers"
• Purpose: Text/sentence embeddings (not generation or classification)

Quick way to find models:

https://huggingface.co/models?library=sentence-transformers&sort=downloads

Recommended custom models:

• BAAI/bge-large-en-v1.5 - Excellent for code (1024d)
• thenlper/gte-large - High quality, multilingual (1024d)
• intfloat/e5-large-v2 - Strong general purpose (1024d)
• sentence-transformers/multi-qa-mpnet-base-dot-v1 - Great for Q&A (768d)

⚠️ Incompatible models (will error):

• GPT, LLaMA, Mistral (text generation)
• BERT classification models
• Any model without sentence-transformers support

AnyRAG validates models automatically and provides clear error messages for incompatible models.

🗂️ Multi-Index Support (Pro)

Pro tier users can create multiple indices with different embedding models to organize content by project, language, or use case.

Quick Start

1. Create an index: Ctrl+Shift+P → AnyRAG Pilot: Create Index
2. Switch indices: Click the database icon in the status bar (bottom right)
3. Manage indices: Actions available in the index switcher (rename, delete)

Your active index (shown in status bar) is automatically used by all commands and searches. Changes persist across sessions.

Example Use Cases

Different embedding models:

• code-specialized - Uses BAAI/bge-large-en-v1.5 for code
• docs-quality - Uses all-mpnet-base-v2 for documentation
• fast-search - Uses all-MiniLM-L6-v2 for quick searches

Separate projects/clients:

• client-acme - Acme Corp project
• client-beta - Beta Inc project
• personal-projects - Your side projects

Each index has its own embedding model and completely separate content.

🗂️ Managing Sources & Tags

Every time you index content (workspace, GitHub repo, file, or chat), AnyRAG creates a source - a logical grouping of that content. Sources can be organized with tags and activated/deactivated to control what's searched.

Viewing Sources

Ctrl+Shift+P → AnyRAG Pilot: Show Indexed Sources

Displays all indexed content with:

• Type indicators: 📁 Folder, 📦 Repo, 📄 File
• Active status: Green checkmark for active sources
• Chunk count: Number of indexed segments
• Tags: Organization labels

Source Actions

Click any source to:

• Add/Remove Tags - Organize with labels like docs, python, important
• Activate/Deactivate - Control which sources are searched
• Rename - Change chat conversation names
• Remove - Delete source and all its data

Active vs Inactive Sources

Only active sources are included in searches. This lets you:

• Focus searches on relevant content
• Switch contexts quickly using tags
• Keep indexed content without searching it

Community Tier Limitation:

• Only 1 source can be active at a time
• Must deactivate current source before activating another
• Pro: Unlimited active sources

Using Tags Effectively

Tag during indexing:

Index Workspace → Auto-tagged: workspace, [folder-name]
Index GitHub Repo → Auto-tagged: github
Index File → Auto-tagged: file, [filename]

Organize within your project:

• core - Main application code
• dependencies - Third-party libraries
• docs - Documentation files
• tests - Test files
• deprecated - Old code to keep but not search

Community users: Switch between sources using tags (only 1 active at a time)

• Working on code: Activate core tagged sources
• Reading docs: Deactivate core, activate docs
• Need library reference: Activate dependencies

📝 License

MIT License - See LICENSE for details.

🛟 Support

• GitHub Issues: Report bugs & request features
• Pro Support: Upgrade for priority support
• Documentation: Full documentation
• Pro Support: Direct developer access