Agent Memory

Agent Memory is a Visual Studio Code extension that gives GitHub Copilot "Long-Term Memory" by indexing your codebase into a local vector database. It enables Retrieval-Augmented Generation (RAG) directly within VS Code, allowing you to ask questions about your entire project with high accuracy.

Why Agent Memory?

Standard LLM context windows are limited. When you work on large projects, Copilot doesn't know about files you haven't opened or referenced. Agent Memory solves this by:

1. Indexing your code: It parses your source code into semantic chunks (functions, classes) using Tree-sitter.
2. Creating Embeddings: It converts these chunks into vector embeddings locally using @xenova/transformers.
3. Storing Locally: It saves these vectors in a local LanceDB database. No code leaves your machine for indexing.
4. Retrieving Context: When you chat with @memory, it semantically searches your codebase and provides the most relevant snippets to the LLM.

Features

• Smart Chunking: Uses AST (Abstract Syntax Tree) parsing to split code by logical units (functions, classes) rather than arbitrary lines. Supports TypeScript, Python, C++, Go, Rust, Java, and more.
• Local & Private: All embeddings and vector data are generated and stored locally on your machine.
• Seamless Integration: Works directly with GitHub Copilot Chat via the @memory participant.
• Memory Management:
  • "Memory Files" View: See exactly what files are indexed, grouped by folder.
  • Visual Inspection: View the actual text chunks and vector data generated for any file.
  • Easy Management: Add or remove files/folders via context menus.

Usage

1. Indexing Code

You can index individual files or entire folders:

• Right-click a file/folder in the Explorer and select "Add File/Folder to Memory Index".
• Open a file and run the command "Add File to Memory Index" from the Command Palette.

2. Chatting with Memory

Open GitHub Copilot Chat and use the @memory participant:

@memory How does the authentication service work?
@memory Where is the user validation logic located?
@memory Explain the relationship between the ChunkingService and MemoryService.

The extension will search your indexed code, retrieve relevant chunks, and use them to answer your question.

3. Managing Memory

Click the Agent Memory icon in the Activity Bar (database icon) to open the Indexed Files view.

• View Chunks: Right-click a file -> View Chunks to see how your code was split.
• View Vectors: Right-click a file -> View Vectors to inspect the raw vector data.
• Delete: Remove files or folders from the index.

Configuration

You can customize the extension in VS Code Settings (Ctrl+,):

• agentMemory.storageScope:
  • workspace (Default): Creates a separate database for each workspace.
  • global: Shares one database across all VS Code instances.
• agentMemory.indexing.maxFileSize: Max file size to index (default: 1MB).
• agentMemory.indexing.excludePatterns: Glob patterns to ignore (e.g., node_modules, dist).
• agentMemory.indexing.includeExtensions: File types to include when indexing folders.

Development Guide

If you want to contribute or modify the extension:

Prerequisites

• Node.js (v18 or higher)
• Visual Studio Code

Setup

1. Clone the repository.
2. Install dependencies:

   npm install
   

3. Compile the extension:

   npm run compile
   

   Note: This step copies the required tree-sitter.wasm file to the output directory.

Running & Debugging

1. Open the project in VS Code.
2. Press F5 to start the "Extension Development Host".
3. In the new window, open a folder and test the commands.

Testing

Run the test suite:

npm test

Release & Changelog

This project uses Conventional Commits and standard-version to automate versioning and changelog generation.

1. Commit Changes: Ensure your commit messages follow the conventional format:

   git commit -m "feat: add new chunking strategy"
   git commit -m "fix: resolve indexing timeout"
   

   A git hook will verify your commit message format.

2. Create Release: Run the release script to bump version and update CHANGELOG.md:

   npm run release
   

   This will:

   • Bump the version in package.json.
   • Update CHANGELOG.md with commits since the last release.
   • Commit the changes and tag the release.

3. Push: This will trigger the GitHub Action to publish to the marketplace:

   git push --follow-tags origin main
   

Publishing to VS Code Marketplace

Automated Publishing (via GitHub Actions):

1. Create a Personal Access Token (PAT) from Azure DevOps:
   • Click on your profile → Security → Personal access tokens.
   • Create a new token with Marketplace (Manage) scope.
2. Add the token to your GitHub repository secrets:
   • Go to Settings → Secrets and variables → Actions.
   • Add a new secret named VSCE_PAT with your token.
3. Push a tag (via npm run release) to trigger automatic publishing.

Manual Publishing:

npx @vscode/vsce publish

Packaging

To create a VSIX file for manual installation:

1. Run the package command using npx:

   npx @vscode/vsce package
   

   This will generate an agent-memory-0.0.1.vsix file in the project root.

2. To install the VSIX:

   • Open VS Code Extensions view.
   • Click the "..." menu.
   • Select "Install from VSIX...".

License

MIT