🧠 Pi SuperMemory Extension

Persistent memory for Pi - the AI coding agent that remembers context across sessions using SuperMemory.

Note: This extension is compatible with SuperMemory SDK v2. API parameters have changed from earlier versions (containerTag instead of containerTags, limit instead of topK, q instead of context, forget instead of delete).

Overview

This extension gives Pi long-term memory capabilities, allowing it to:

• Remember user preferences across all projects (coding style, favorite tools, etc.)
• Maintain project context between sessions (architecture decisions, pending tasks)
• Learn from conversations and recall relevant information when needed
• Build a user profile that accumulates facts about you over time

Features

🔄 Automatic Context Injection

• Relevant memories are automatically injected into prompts based on semantic similarity
• No manual searching required - memories surface when they're needed

🎯 Two Memory Scopes

• User scope: Cross-project knowledge that follows you everywhere
• Project scope: Project-specific knowledge tied to the current directory

🤖 LLM Tool Access

The supermemory tool gives the AI direct access to:

• add - Store new memories with type classification
• search - Find relevant memories by query
• profile - View accumulated user profile
• list - Browse recent memories
• forget - Delete specific memories

⌨️ Slash Commands

• /remember <text> - Quick-save something to memory
• /recall <query> - Search your memories
• /memory-profile - View your user profile

💾 Automatic Conversation Capture

Conversations are automatically stored (when substantial) for future reference.

Installation

Prerequisites

1. Get a SuperMemory API key:

   • Sign up at supermemory.ai
   • Generate an API key from your dashboard

2. Set environment variables:

   export SUPERMEMORY_API_KEY="your-api-key-here"
   # Optional: customize container prefix (default: "pi")
   export SUPERMEMORY_CONTAINER="myorg"

Install the Extension

# In your Pi extensions directory (usually ~/.pi/agent/extensions/)
cd ~/.pi/agent/extensions/

# Clone or copy this extension
git clone https://github.com/dan/pi-supermemory.git supermemory

# Install dependencies
cd supermemory
npm install

Verify Installation

Start Pi and look for the 🧠 indicator in the status bar, or check that no "SUPERMEMORY_API_KEY not set" warning appears.

Usage

Let the AI Remember Things

Simply tell Pi to remember something:

"Remember that I prefer 2-space indentation in TypeScript" "Save this for later - the database connection string is..." "Remember this project uses a monorepo structure"

The AI will use the supermemory tool to store this information with the appropriate scope and type.

Manual Commands

# Quick save
/remember I always use pnpm instead of npm

# Search memories
/recall database schema

# View your profile
/memory-profile

Memory Types

When storing memories, they can be classified as:

| Type | Description | |------|-------------| | preference | Personal preferences (coding style, tools, etc.) | | project-config | Project configuration details | | architecture | System design and architecture decisions | | error-solution | Solutions to problems you've encountered | | learned-pattern | General patterns and insights | | conversation | Automatically captured conversations |

Memory Scopes

• user: Available across all projects (good for personal preferences)
• project: Only available in the current project directory (default)

Configuration

Environment Variables

| Variable | Required | Default | Description | |----------|----------|---------|-------------| | SUPERMEMORY_API_KEY | Yes | - | Your SuperMemory API key | | SUPERMEMORY_CONTAINER | No | pi | Prefix for container tags |

How Container Tags Work

The extension uses container tags to organize memories:

• User memories: {SUPERMEMORY_CONTAINER}_user_{username}
• Project memories: {SUPERMEMORY_CONTAINER}_project_{project_name}

This keeps different users' and projects' memories separate.

API Reference

Tool: supermemory

The LLM can call this tool with the following parameters:

{
  mode: "add" | "search" | "profile" | "list" | "forget" | "help",
  content?: string,      // For "add" mode
  query?: string,        // For "search" or "profile" modes
  type?: MemoryType,     // For "add" mode
  scope?: "user" | "project",  // For "add", "search", "list" modes
  memoryId?: string,     // For "forget" mode
  limit?: number         // For "search" and "list" modes
}

Note: The SuperMemory SDK v2 API uses containerTag (singular), limit (instead of topK), and q (instead of context). This extension has been updated to match the latest SDK.

Example Tool Calls

// Store a preference
{ mode: "add", content: "Prefers dark mode themes", type: "preference", scope: "user" }

// Search for relevant memories
{ mode: "search", query: "database connection", scope: "project", limit: 5 }

// View user profile
{ mode: "profile", query: "coding preferences" }

// List recent memories
{ mode: "list", scope: "project", limit: 10 }

// Forget a memory
{ mode: "forget", memoryId: "mem_abc123", scope: "project" }

Architecture

The extension works by:

1. Before each agent turn: Searches SuperMemory for context relevant to the user's prompt, injects matching memories into the system prompt

2. After each agent turn: Captures the conversation text and stores it in project scope for future reference

3. During tool execution: Provides the LLM with tools to actively manage memory (add, search, delete)

4. On session start: Shows a 🧠 status indicator when memory is active

Development

# Clone the repo
git clone https://github.com/dan/pi-supermemory.git
cd pi-supermemory

# Install dependencies
npm install

# Link for local development
npm link

# In Pi extensions directory
npm link pi-supermemory

Troubleshooting

"SUPERMEMORY_API_KEY not set - extension disabled"

• Make sure you've set the SUPERMEMORY_API_KEY environment variable
• Restart Pi after setting environment variables

Memories not appearing

• Check that the SuperMemory API key is valid
• Verify network connectivity to supermemory.ai
• Check Pi's console for error messages

Wrong project scope

• The project scope is derived from the current working directory name
• Make sure you're in the correct project directory

Related

• Pi Coding Agent - The AI coding agent this extension is for
• SuperMemory - The memory service powering this extension

License

MIT

Contributing

Contributions welcome! Please open an issue or PR on GitHub.