Internet of Agents

An open-source framework by OpenBMB for collaborative AI agents, enabling heterogeneous agents to team up autonomously through internet-like architecture with async task execution.

Overview#

IoA (Internet of Agents) is an open-source collaborative AI agent framework developed by the OpenBMB team. Inspired by internet architecture, it connects heterogeneous, distributed AI agents, enabling them to autonomously form teams and collaborate on complex tasks.

Key Features#

Architecture Capabilities#

Internet-like Connectivity: Architecture design supporting cross-device, cross-environment agent interconnection
Heterogeneous Agent Integration: Supports integration of agents from different sources and architectures (verified support for AutoGPT, Open Interpreter, ReAct Agent)
Asynchronous Task Execution: Async processing mechanism for improved system concurrency and efficiency

Collaboration Mechanism#

Autonomous Nested Team Collaboration: Agents can autonomously form teams and sub-teams based on task requirements
Adaptive Dialogue Flow: Automatically manages dialogue structure between agents, dynamically adjusting interaction flow based on task progress

Extensibility#

Architecture design allows developers to easily add new types of agents
Instant messaging-style architecture ensures clear agent communication structure

Architecture Components#

Component	Description
IoA Server	Core server handling scheduling, communication management, and state maintenance
IoA Client	Client encapsulating specific Agent logic, connecting to Server
Server Frontend	Web UI for monitoring and interaction
Milvus	Vector database for semantic retrieval and memory functions

Installation & Deployment#

Requirements#

Docker (Required)
Docker Compose (Recommended)

Quick Start#

Clone repository

git clone git@github.com:OpenBMB/IoA.git
cd IoA

Pull core component images

docker pull weize/ioa-server:latest
docker pull weize/ioa-client:latest
docker pull weize/ioa-server-frontend:latest

Pull agent images (as needed)

docker pull weize/react-agent:latest
docker pull weize/autogpt:latest
docker pull weize/open-interpreter:latest

Start Milvus service

docker network create agent_network
docker-compose -f dockerfiles/compose/milvus.yaml up -d

Configure environment and start

cd dockerfiles/compose/
cp .env_template .env
# Edit .env file with OPENAI_API_KEY and other required values
docker-compose -f dockerfiles/compose/open_instruction.yaml up

Usage Example#

HTTP Request#

import requests

goal = "I want to know the annual revenue of Microsoft from 2014 to 2020..."

response = requests.post(
    "http://127.0.0.1:5050/launch_goal",
    json={
        "goal": goal,
        "max_turns": 20,
        "team_member_names": ["AutoGPT", "Open Interpreter"],
    },
)

API Interface#

POST /launch_goal: Core endpoint for launching collaborative tasks
- goal (string): Task description
- max_turns (int): Maximum turns
- team_member_names (list[string]): Team member names

Use Cases#

Complex data analysis tasks (cross-timeframe data retrieval and visualization)
Multi-tool tasks (code writing + execution and debugging)
Distributed AI collaboration research and development