What is AgentScope?

AgentScope is a production-ready, easy-to-use agent framework with essential abstractions that work with rising model capability and built-in support for finetuning.

We design for increasingly agentic LLMs. Our approach leverages the models' reasoning and tool use abilities rather than constraining them with strict prompts and opinionated orchestrations.

Why use AgentScope?

• Simple: start building your agents in 5 minutes with built-in ReAct agent, tools, skills, human-in-the-loop steering, memory, planning, realtime voice, evaluation and model finetuning
• Extensible: large number of ecosystem integrations for tools, memory and observability; built-in support for MCP and A2A; message hub for flexible multi-agent orchestration and workflows
• Production-ready: deploy and serve your agents locally, as serverless in the cloud, or on your K8s cluster with built-in OTel support

Requirements

• Node.js: 18.0.0 or higher (20.x recommended)
• Package Manager: pnpm 8.0.0+ (recommended) or npm

Features

• Human-in-the-loop: Built-in support for user confirmation and external execution, allowing seamless human intervention in agent workflows.

• Event System: Multi-agent oriented event-driven architecture for tracking agent lifecycle, tool calls, and streaming responses, designed to be frontend-friendly.

• MCP Support: Full integration with Model Context Protocol via both HTTP and stdio transports, enabling standardized tool communication.

• Skill System: Extensible skill system for adding custom capabilities with automatic schema generation and hot-reloading support.

Quick Start

pnpm install @agentscope-ai/agentscope
# or
# npm install @agentscope-ai/agentscope

A quick example for using AgentScope

import { Agent } from '@agentscope-ai/agentscope/agent';
import { OpenAIModel } from '@agentscope-ai/agentscope/model';
import { Toolkit, Bash, Read, Write } from '@agentscope-ai/agentscope/tool';

// Create a model
const model = new OpenAIModel({
    apiKey: process.env.OPENAI_API_KEY,
    model: 'gpt-4o',
});

// Create a toolkit with built-in tools
const toolkit = new Toolkit({
    tools: [Bash(), Read(), Write()],
    skills: ['/path/to/your/skills'],
    skillDirs: ['/path/to/your/skillDirs'], // Monitoring these directories
});

// Initialize an agent
const agent = new Agent({
    name: 'Friday',
    sysPrompt: 'You are a helpful assistant named Friday.',
    model: model,
    toolkit: toolkit,
});

// Use the agent
const res = await agent.reply();
console.log(res.text);

// Or, get streaming events from the agent reply
let name = '';
for await (const event of agent.replyStream({})) {
    switch (event.type) {
        case EventType.RUN_STARTED:
            name = event.name;
            break;
        case EventType.TEXT_BLOCK_START:
            process.stdout.write(`${name}: `);
            break;
        case EventType.TEXT_BLOCK_DELTA:
            process.stdout.write(event.delta);
            break;
        case EventType.TEXT_BLOCK_END:
            process.stdout.write('\n\n');
            break;
    }
}

Project Structure

This is a monorepo containing:

• packages/agentscope: Core AgentScope library
• app/friday: Example Electron application built with AgentScope

License

AgentScope is released under Apache License 2.0.

Contributors

All thanks to our contributors: