Getting Started with AgentKits

What is AgentKits?

AgentKits is a modular framework for building AI agents with real-world skills. Instead of writing everything from scratch, you compose agents from skill kits — pre-built modules for memory, browser automation, filesystem operations, and more.

Think of it like building blocks for AI agents. Each kit handles one capability well, and they snap together cleanly.

Installation

AgentKits is distributed as npm packages. Install the core framework and any kits you need:

npm install @agentkits/core @agentkits/memory @agentkits/browser

Or use a single meta-package that includes all official kits:

npm install agentkits

Your First Agent

Here’s a minimal agent with memory capabilities:

import { Agent } from '@agentkits/core';
import { MemoryKit } from '@agentkits/memory';

const agent = new Agent({
  name: 'my-agent',
  kits: [new MemoryKit()],
});

await agent.run('Remember that the user prefers dark mode');
const result = await agent.run('What are the user preferences?');
console.log(result);

The agent now has persistent memory. It can store facts, recall them later, and use them in context.

Available Kits

Memory Kit

Gives your agent persistent memory using vector embeddings. Store facts, conversations, and context that persists across sessions.

Browser Kit

Lets your agent interact with web pages — navigate, click, fill forms, extract data. Built on Puppeteer with a high-level API designed for agents.

Filesystem Kit

Read, write, and manage files and directories. Includes safety guards to prevent accidental data loss and sandbox mode for untrusted operations.

Key Concepts

Kit Composition

Agents gain capabilities by composing kits. Each kit registers tools that the agent can call during execution:

const agent = new Agent({
  kits: [
    new MemoryKit({ maxEntries: 1000 }),
    new BrowserKit({ headless: true }),
    new FilesystemKit({ sandbox: './workspace' }),
  ],
});

Tool Registration

Each kit exposes tools with typed schemas. The agent’s LLM backbone sees these tools and decides when to use them based on the task.

Safety by Default

All kits implement safety checks. The Browser Kit won’t navigate to file:// URLs. The Filesystem Kit respects sandbox boundaries. The Memory Kit encrypts sensitive entries at rest.

Next Steps

• Read the full documentation for detailed API reference
• Browse example agents for real-world patterns
• Join the conversation on GitHub

---

Need help? Open an issue on GitHub or email support@agentkits.net.