Memory

Imagine if your coding agents could reuse the project decisions you have already made.

The Memory service allows you to store actionable insights, coding patterns, and project guidelines. Once stored, this knowledge is available to your AI coding agents (via MCP) and to you directly via the CLI, ensuring consistency across your development workflow.

Before using Memory, ensure you have:

• Node.js 20.20.0 or higher installed
• AI DevKit CLI installed: npm install -g ai-devkit or use npx ai-devkit@latest
• For MCP usage: A compatible AI coding agent or environment (Cursor, Claude Code, etc.)

You can interact with Memory in three ways:

1. Through your AI Assistant (Recommended): Connect via MCP so your AI can automatically search for relevant context and save new rules as you work.
2. Through the CLI: Manually store or retrieve knowledge directly from your terminal—perfect for quick lookups or scripting.
3. Through Skills: Install the memory skill to teach your AI agent how to use CLI commands when MCP is unavailable.

This is the most powerful way to use Memory. Your AI (Cursor, Claude, etc.) gains "tools" to save and retrieve information naturally.

Add the server to your MCP configuration file:

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["@ai-devkit/memory"]
    }
  }
}

Once connected, you can talk to your AI naturally:

Storing Knowledge:

"We just decided that all API responses must handle BigInt serialization. Please save this rule to memory with the tag #backend."

You can also ask your AI agent to store knowledge directly when the memory skill is available.

Example:

Remember that we always use BigInt.toString() before sending API responses containing large numbers.

The AI will prompt you for a title and tags, then save the knowledge to memory.

Retrieving Knowledge:

"I'm building a new endpoint. Check memory for any API standards I need to follow."

The AI will intelligently rank results based on your current task and available tags.

You don't need an AI agent to use Memory. The ai-devkit CLI has built-in commands to manage your knowledge base.

Found a solution to a tricky bug? Save it immediately:

ai-devkit memory store \
  --title "Fix: Docker connection refused on M1 Mac" \
  --content "Enable 'Use Rosetta for x86/amd64 emulation' in Docker Desktop settings." \
  --tags "docker,mac,infra" \
  --scope "global"

Parameter	Required	Description
--title	Yes	A short, descriptive title
--content	Yes	The detailed knowledge to store
--tags	No	Comma-separated tags for categorization
--scope	No	global (default), project:<name>, or repo:<org/repo>

💡 Tip: Run ai-devkit memory store --help to see all available options.

Need to recall a specific command or pattern? Search for it:

ai-devkit memory search --query "docker m1"

Output:

{
  "query": "docker m1",
  "results": [
    {
      "id": "...",
      "title": "Fix: Docker connection refused on M1 Mac",
      "content": "Enable 'Use Rosetta for x86/amd64 emulation' in Docker Desktop settings.",
      "tags": ["docker", "mac", "infra"],
      "scope": "global",
      "score": 5.2
    }
  ]
}

Useful options:

• --limit <n> to control how many results are returned
• --scope <scope> to filter results to one scope
• --tags <tags> to boost matches using context tags
• --table to print a compact table with id, title, and scope

Note: If no results are found, the results array is empty.

If an existing memory item is still useful but needs corrections or more detail, update it by ID:

ai-devkit memory update \
  --id "<memory-id>" \
  --title "Fix: Docker connection refused on Apple Silicon" \
  --content "Enable 'Use Rosetta for x86/amd64 emulation' in Docker Desktop settings, then restart Docker Desktop."

You can also replace tags or scope during an update:

ai-devkit memory update \
  --id "<memory-id>" \
  --tags "docker,mac,infra" \
  --scope "global"

If MCP is not available in your environment, you can install the memory skill to teach your AI agent how to use memory via CLI commands.

ai-devkit skill add codeaholicguy/ai-devkit memory

This installs the memory skill into your project's skill directory (e.g., .cursor/skills/memory/).

Once installed, your AI agent will have access to:

• Detailed CLI command reference for memory store and memory search
• Parameter documentation with examples for all options
• Best practices for crafting titles, content, and tags
• Troubleshooting tips for common issues

The memory skill is ideal when:

• MCP is not configured or unavailable
• Your AI agent needs detailed command syntax reference
• You want consistent memory usage patterns across your team

Tip: The skill works with all skill-capable AI environments: Cursor, Claude Code, GitHub Copilot, Codex, opencode, Antigravity, Junie, Cline, Devin, Pi, Kilo Code, and Roo Code.

To keep your memory effective, use Tags and Scopes.

Categorize your entries so they trigger in the right context.

• ["typescript", "react"] -> For frontend rules.
• ["deployment", "ci"] -> For DevOps procedures.

Control where your knowledge applies:

Scope	CLI Flag	Description
Global	--scope global	Applies to all your projects (default)
Project	--scope project:my-app	Specific to a project
Repo	--scope repo:org/repo	Specific to a git repository

Example:

ai-devkit memory store \
  --title "Use pnpm for this monorepo" \
  --content "This project uses pnpm workspaces. Always use 'pnpm' instead of 'npm'." \
  --scope "repo:myorg/my-monorepo"

Note: AI agents automatically prioritize knowledge matching the scope of your current working directory.

Your memory is 100% local.

By default, Memory stores data in a local SQLite database at ~/.ai-devkit/memory.db.

If you use the ai-devkit CLI, you can override that location in your .ai-devkit.json file with memory.path. When memory.path is set, CLI commands use that path instead. Absolute paths are used as-is, and relative paths are resolved from the directory that contains .ai-devkit.json.

{
  "memory": {
    "path": ".ai-devkit/project-memory.db"
  }
}

No data is sent to the cloud, ensuring your proprietary coding patterns remain private.

Because the database is just a local file, you can copy it to another machine and keep using the same memory there.

Plugins can also read the configured memory database path through runtime.getMemoryDbPath(). This is useful for plugin dashboards and local tools that need to inspect the same memory database as the CLI.

• Skills: Learn how to create reusable skill templates
• Plugins: Add optional CLI commands that can use the memory runtime
• Getting Started: New to AI DevKit? Start here