Agent Memory
A VS Code extension that implements Claude's memory tool for Language Model agents. This extension enables AI agents to store and retrieve information across conversations through a persistent memory file system.
Features
The extension provides a memory tool that AI agents (like GitHub Copilot) can use to:
- View: Read directory contents or file contents with optional line ranges
- Create: Create or overwrite files in the memory directory
- String Replace: Replace specific text within files
- Insert: Insert text at specific line numbers
- Delete: Remove files or directories
- Rename: Rename or move files and directories
All operations are scoped to the /memories directory with path traversal protection for security.
Storage Backends
Choose between three storage options via settings (agentMemory.storageBackend):
- In-Memory (default): Ephemeral storage per workspace, cleared when VS Code closes
- Disk: Persistent storage in
.vscode/memory directory within your workspace
- Secret: Encrypted persistent storage using VS Code's Secret Storage API - ideal for sensitive information
Visual Interface
Memory Files View
- Tree view showing all memory files and directories
- Display file size and last access time
- Click to open and view file contents
- Pin/Unpin Files: Pin important files to keep them at the top of the tree view
- Automatic refresh when memory operations occur
- Context menu actions: pin/unpin, delete, save as markdown
Activity Log View
- Real-time log of all memory operations
- Shows command type, path, timestamp, and success/failure
- Clear logs with one click
- Visual indicators for successful vs failed operations
File Auto-Sync
When configured via the agentMemory.autoSyncToFile setting, the extension automatically synchronizes all memory files to a specified file in your workspace (e.g., AGENTS.md or .github/copilot/memory.instructions.md). This provides a centralized, readable view of all agent memory in a format that's easy to review and version control.
Features:
- Specify any relative file path for syncing (e.g.,
AGENTS.md, .github/copilot/memory.instructions.md)
- Memory content is wrapped in
<memories hint="Manage via memory tool">...</memories> tags
- Each file is enclosed in
<memory path="/memories/...">...</memory> tags
- Automatically updates when memory files are created, modified, or deleted
- Preserves existing file content outside the memory section
- If the target file doesn't exist, one will be created automatically
- For
.instructions.md files, adds frontmatter with applyTo: ** directive
Example AGENTS.md output:
# My Project Documentation
<memories hint="Manage via memory tool">
<memory path="/memories/preferences.txt">
- Prefers TypeScript for new projects
- Uses ESLint for code quality
</memory>
<memory path="/memories/context.txt">
Project is a VS Code extension for AI agent memory management.
</memory>
</memories>
Example .instructions.md output:
---
applyTo: **
---
<memories hint="Manage via memory tool">
<memory path="/memories/preferences.txt">
- Prefers TypeScript for new projects
- Uses ESLint for code quality
</memory>
</memories>
Usage
For AI Agents
The memory tool is automatically available to AI agents in your workspace. Agents can reference it using the tool name memory or via the #memory tool reference.
Example agent interaction:
User: "Remember that I prefer using TypeScript for new projects"
Agent: I'll store that preference in memory.
[calls memory tool with command: "create", path: "/memories/preferences.txt", file_text: "- Prefers TypeScript for new projects"]
Commands
- Agent Memory: Refresh Memory Files - Manually refresh the memory files view
- Agent Memory: Clear Activity Logs - Clear the activity log
- Agent Memory: Clear All Memory Files - Delete all memory files and directories
- Agent Memory: Delete Memory File - Delete a specific memory file (available in context menu)
- Agent Memory: Open Memory File - Open a memory file for viewing
- Agent Memory: Pin File - Pin a file to keep it at the top of the tree view
- Agent Memory: Unpin File - Remove pin from a file
- Agent Memory: Save as Markdown - Export a memory file as a markdown document
Settings
agentMemory.storageBackend: Choose between memory (default), disk, or secret storageagentMemory.autoSyncToFile: Relative file path to automatically sync all memory files to (e.g., AGENTS.md or .github/copilot/memory.instructions.md). Leave empty to disable auto-sync. (default: "" - disabled)
View Command
{
"command": "view",
"path": "/memories",
"view_range": [1, 10] // Optional
}
Create Command
{
"command": "create",
"path": "/memories/notes.txt",
"file_text": "Content here"
}
String Replace Command
{
"command": "str_replace",
"path": "/memories/file.txt",
"old_str": "old text",
"new_str": "new text"
}
Insert Command
{
"command": "insert",
"path": "/memories/file.txt",
"insert_line": 5,
"insert_text": "New line text"
}
Delete Command
{
"command": "delete",
"path": "/memories/file.txt"
}
Rename Command
{
"command": "rename",
"old_path": "/memories/old.txt",
"new_path": "/memories/new.txt"
}
Security
- All paths are validated to prevent directory traversal attacks
- Memory operations are restricted to the
/memories directory only
- Path normalization prevents access outside the memory directory
Secret Storage Backend
The secret storage backend uses VS Code's built-in Secret Storage API, which:
- Encrypts memory files using the platform's secure credential storage (Keychain on macOS, Credential Vault on Windows, Secret Service on Linux)
- Stores file metadata separately for efficient listing operations
- Provides workspace-scoped secrets that persist across VS Code sessions
- Ideal for storing sensitive information like API keys, tokens, or personal notes
Running the Extension
- Run
npm install in terminal to install dependencies
- Run the
Run Extension target in the Debug View. This will:
- Start a task
npm: watch to compile the code
- Run the extension in a new VS Code window
- The memory tool will be registered and available to any chat participant
Architecture
- Storage Layer: Abstracted storage interface supporting multiple backends (in-memory, disk, and secret)
- Activity Logger: Tracks all memory operations with metadata
- Tree Providers: Real-time UI updates for files and activity logs
- Tool Integration: Native VS Code Language Model Tool API
See tools.ts for the memory tool implementation, storage.ts for storage backends, and views.ts for the UI components.
Requirements
- VS Code 1.105.0 or higher
- A workspace folder must be open to use memory features
Credits
Based on Claude's memory tool specification from Anthropic and the VS Code Language Model Tools API.
License
MIT
