​

Deploying Multi-Agent Systems as APIs

Praison Labs allows you to deploy sophisticated multi-agent systems as RESTful APIs, enabling seamless integration with various applications and services. This guide covers different approaches to deploying multi-agent systems.

​

Quick Start

pip install "praisonaiagents[api]>=0.0.81"

export OPENAI_API_KEY="your_api_key"

from praisonaiagents import Agent, Agents, Tools

research_agent = Agent(name="Research", instructions="You are a research agent to search internet about AI 2024", tools=[Tools.internet_search])
summarise_agent = Agent(name="Summarise", instructions="You are a summarize agent to summarise in points")
agents = Agents(agents=[research_agent, summarise_agent])
agents.launch(path="/agents", port=3030)

python multi-agents-api.py

​

Making API Requests

Once your multi-agent system is deployed, you can make POST requests to interact with it:

curl -X POST http://localhost:3030/agents \
  -H "Content-Type: application/json" \
  -d '{"message": "What are the latest developments in AI in 2024?"}'

The response will include the collaborative output from both the research and summarization agents:

{
  "response": "# Latest AI Developments in 2024\n\n- Multimodal AI models have become mainstream, combining text, image, audio, and video understanding\n- Significant advancements in AI reasoning capabilities with models like GPT-4o and Claude 3\n- Increased focus on AI alignment and safety research\n- Emergence of specialized AI agents for specific domains\n- Growth in open-source AI models and frameworks\n- Regulatory frameworks for AI being established globally"
}

​

Multiple Agent Groups

You can deploy multiple agent groups on the same server, each with its own endpoint:

from praisonaiagents import Agent, Agents, Tools

research_agent = Agent(name="Research", instructions="You are a research agent to search internet about AI 2024", tools=[Tools.internet_search])
summarise_agent = Agent(name="Summarise", instructions="You are a summarize agent to summarise in points")
agents = Agents(agents=[research_agent, summarise_agent])
agents2 = Agents(agents=[research_agent])
agents.launch(path="/agents", port=3030)
agents2.launch(path="/agents2", port=3030)

With this setup, you can access:

• The full agent group at http://localhost:3030/agents
• Just the research agent at http://localhost:3030/agents2

​

Independent Multi-Agent Deployment

You can also deploy multiple independent agents on the same server:

from praisonaiagents import Agent

weather_agent = Agent(
    instructions="""You are a weather agent that can provide weather information for a given city.""",
    llm="gpt-4o-mini"
)

stock_agent = Agent(
    instructions="""You are a stock market agent that can provide information about stock prices and market trends.""",
    llm="gpt-4o-mini"
)

travel_agent = Agent(
    instructions="""You are a travel agent that can provide recommendations for destinations, hotels, and activities.""",
    llm="gpt-4o-mini"
)

weather_agent.launch(path="/weather", port=3030)
stock_agent.launch(path="/stock", port=3030)
travel_agent.launch(path="/travel", port=3030) 

​

Production Deployment Options

For production environments, consider the following deployment options:

​

Docker Deployment

FROM python:3.11-slim

WORKDIR /app

COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

COPY . .

EXPOSE 3030

CMD ["python", "multi-agents-api.py"]

praisonaiagents[api]>=0.0.81

docker build -t praisonai-multi-agent .
docker run -p 3030:3030 -e OPENAI_API_KEY=your_api_key praisonai-multi-agent

​

Cloud Deployment

For detailed cloud deployment instructions, refer to:

• AWS Deployment Guide
• Google Cloud Deployment Guide

​

Scaling Multi-Agent Systems

When deploying multi-agent systems to production, consider these scaling strategies:

1. Horizontal Scaling: Deploy multiple instances behind a load balancer
2. Vertical Scaling: Allocate more CPU and memory resources for complex agent interactions
3. Caching: Implement response caching for frequently asked questions
4. Asynchronous Processing: Use message queues for handling long-running agent tasks

​

API Configuration Options

When launching your multi-agent system as an API, you can customize various parameters:

agents.launch(
    path="/custom-endpoint",  # API endpoint path
    port=8080,                # Port number
    host="0.0.0.0",           # Host address (0.0.0.0 for external access)
    debug=True,               # Enable debug mode
    cors_origins=["*"],       # CORS configuration
    api_key="your-api-key"    # Optional API key for authentication
)

​

Securing Your Multi-Agent API

For production deployments, consider implementing:

1. API Key Authentication: Require API keys for all requests
2. Rate Limiting: Limit the number of requests per client
3. HTTPS: Use SSL/TLS certificates for encrypted communication
4. Input Validation: Validate all input data before processing
5. Output Filtering: Implement content filtering for agent responses

​

Monitoring and Logging

For production environments, consider:

1. Centralized Logging: Collect logs from all agents in a central location
2. Performance Metrics: Track response times and resource usage
3. Error Tracking: Monitor and alert on agent failures
4. Usage Analytics: Track which agents are used most frequently

​

Features