agent-kit

TypeScript-first library for building stateful, persistent AI agents.

Live Playground | Blog Post | npm | Docs

---

Why agent-kit?

• Persistent memory across sessions — SQLite or PostgreSQL. Restart your process; the agent still remembers.
• Simple tool system — define a tool in 5 lines with Tool.create. No decorators, no class inheritance.
• Multi-agent coordination — 4 strategies: Sequential, Parallel, Debate, Hierarchical.
• Built-in observability — subscribe to tool:start, tool:end, memory:save events. No paid add-on.
• TypeScript-first — strict types throughout. Your editor autocompletes everything.

---

4 Core Concepts

---

Installation

npm install @avee1234/agent-kit

Or scaffold a new project:

npx @avee1234/agent-kit init my-agent

---

Quick Start

import { Agent, Tool, Memory } from '@avee1234/agent-kit';

const searchTool = Tool.create({
  name: 'web_search',
  description: 'Search the web',
  parameters: { query: { type: 'string' } },
  execute: async ({ query }) => fetch(`https://api.search.com?q=${query}`).then(r => r.json()),
});

const agent = new Agent({
  name: 'research-assistant',
  model: { provider: 'ollama', model: 'llama3' },
  memory: new Memory({ store: 'sqlite' }),
  tools: [searchTool],
  system: 'You are a research assistant.',
});

const response = await agent.chat('Find recent papers on transformers');

No model config required — ships with a built-in MockAdapter for development and testing without API keys.

---

The "Wow Moment": Memory That Survives Restarts

Kill the process, restart it, same SQLite file — the agent picks up where it left off.

---

Multi-Agent Coordination

const team = new Team({
  agents: [researcher, writer, critic],
  strategy: 'debate',
  maxRounds: 3,
});
const result = await team.run('What is the best database for embeddings?');

---

Built-In Observability

agent.on('*', (e) => console.log(e.type, e.data, e.latencyMs));

No LangSmith, no third-party service. Just EventEmitter — pipe it to whatever you use.

---

Where Each Tool Shines

---

Model Configuration

// Mock (zero config, built-in — great for testing)
const agent = new Agent({ name: 'test' });

// Ollama (local models)
const agent = new Agent({ name: 'local', model: { provider: 'ollama', model: 'llama3' } });

// Google AI Studio (Gemini)
const agent = new Agent({
  name: 'gemini',
  model: new OpenAICompatibleAdapter({
    baseURL: 'https://generativelanguage.googleapis.com/v1beta/openai',
    model: 'gemini-2.0-flash',
    apiKey: process.env.GOOGLE_AI_API_KEY,
  }),
});

// Any OpenAI-compatible endpoint (Together, OpenRouter, Groq, etc.)
const agent = new Agent({
  name: 'agent',
  model: new OpenAICompatibleAdapter({ baseURL: '...', model: '...', apiKey: '...' }),
});

---

Try It Live

www.abhi-agent-kit.space

4 pre-built agents with real-time tool execution, persistent memory, and live event streaming. Powered by Gemini Flash on Google Cloud Run.

---

What You Can Build

• Travel planner — destination search + weather + flight/hotel booking + budget calculator
• Research assistant — web search + note-taking + session memory
• Customer support — order lookup + conversation history
• Code reviewer — security analysis + style checking
• Data analyst — SQL queries + chart generation + persistent findings
• Personal assistant — calendar + email + long-term preferences

---

Read More

• Blog Post — the full story of why and how I built agent-kit
• Documentation — API reference, guides, and examples
• npm — npm install @avee1234/agent-kit

Contributing

Bug reports and pull requests welcome. Open an issue to discuss changes before submitting a PR.

License

MIT