Multi-Agent AI Platform

This project is a multi-agent AI platform consisting of three main agents: Agent Zero, Agent Alpha, and a Researcher Agent. The entire platform is orchestrated using Docker Compose, providing a scalable and production-ready environment.

Architecture

The platform is designed around a central UI provided by Agent Zero. Agent Zero acts as the primary interface and orchestrator, capable of delegating tasks to specialized backend agents like Agent Alpha and various Researcher Agents.

• Agent Zero (agent_zero): The main user-facing application (UI on port 8081), providing tools to manage tasks, and an underlying agent capable of orchestrating with other backend services.
• Agent Alpha Frontend (agent_alpha_frontend): A UI for the Agent Alpha system (UI on port 8082).
• Agent Alpha (agent_alpha): A backend agent system responsible for creating and running intelligent agents for various tasks.
• Researcher Agent - WebWalker (research_agent_webwalker_ui): A backend agent specialized in web traversal research tasks, with a Streamlit UI (UI on port 7860). An internal API service (research_agent_webwalker_api) is also available for Agent Zero.
• Researcher Agent - WebDancer (research_agent_webdancer_ui): A backend agent specialized in autonomous information seeking, with a Gradio UI (UI on port 7861). An internal API service (research_agent_webdancer_api) is also available for Agent Zero.
• Researcher Agent - WebSailor (abusallam/Websailor) (research_agent_websailor_api): An advanced research agent for complex reasoning and information seeking. Interacted with via an internal API by Agent Zero. This service also depends on an SGLang model server (sglang_server_websailor). This version of WebSailor does not have a separate UI.
• LiveKit (livekit): A real-time communication service.
• Redis (redis): An in-memory data store.

All services are connected via an internal Docker network (agents-net), ensuring secure and efficient communication.

Getting Started

Prerequisites

• Docker and Docker Compose installed on your system.
• Git for cloning the repository.

Setup

1. Clone the repository:

   git clone <repository_url>
   cd <repository_directory>

2. Create an environment file: Create a .env file in the root of the project by copying the .env.example file:

   cp .env.example .env

   Update the .env file with your API keys and other configuration values. Ensure you have copied agent_zero_base/.env.example to agent_zero_base/.env if applicable for Agent Zero specific settings, and the root .env.example to .env for platform-wide keys. The root .env file should include:

   OPENAI_API_KEY=your_openai_api_key
   GOOGLE_API_KEY=your_google_api_key      # General Google API Key
   LIVEKIT_API_KEY=your_livekit_api_key
   LIVEKIT_API_SECRET=your_livekit_api_secret
   
   # Required for Researcher Agents (WebWalker, WebDancer)
   DASHSCOPE_API_KEY=your_dashscope_api_key
   OPENAI_MODEL_SERVER=your_openai_compatible_model_server_url # e.g., for local models or specific OpenAI deployments
   GOOGLE_SEARCH_KEY=your_google_custom_search_api_key # Specific for Google Search (Serper, etc.)
   JINA_API_KEY=your_jina_api_key
   
   # Required for WebSailor (abusallam/Websailor)
   WEBSAILOR_SGLANG_MODEL_IDENTIFIER="Qwen/Qwen1.5-7B-Chat" # Example, use desired model
   WEBSAILOR_LLM_LOCAL_PATH_FOR_TOKENIZER="Qwen/Qwen1.5-7B-Chat" # Or path to local model for tokenizer
   # WEBSAILOR_SGLANG_TP_SIZE=1 # Optional
   # WEBSAILOR_MAX_LLM_CALLS=15 # Optional
   # WEBSAILOR_TEMPERATURE=0.6 # Optional
   # WEBSAILOR_TOP_P=0.95 # Optional
   

   (Note: Refer to .env.example for the full list, defaults, and detailed comments for all variables, including those for WebDancer's specific LLM backend configuration.)

Running the Platform

To run the entire platform, use the following Docker Compose command:

docker-compose up --build

This will build the Docker images for all agents and start all the services defined in docker-compose.yml.

Accessing the UIs

Once all the services are running, you can access the UIs in your browser at:

• Agent Zero UI: http://localhost:8081
  • From Agent Zero's settings, you can also find links to the other agent UIs.
• Agent Alpha Frontend UI: http://localhost:8082
• WebWalker UI (Researcher Agent): http://localhost:7860
• WebDancer UI (Researcher Agent): http://localhost:7861

Services

The docker-compose.yml file defines the following key services:

• agent_zero: The main UI application and orchestration agent.
• agent_alpha_frontend: Frontend UI for Agent Alpha.
• agent_alpha: Backend system for Agent Alpha.
• research_agent_webwalker_ui: WebWalker research agent UI.
• research_agent_webwalker_api: WebWalker research agent internal API.
• research_agent_webdancer_ui: WebDancer research agent UI.
• research_agent_webdancer_api: WebDancer research agent internal API.
• sglang_server_websailor: SGLang model server for WebSailor.
• research_agent_websailor_api: WebSailor (abusallam/Websailor) research agent internal API.
• livekit: The real-time communication server.
• redis: The Redis data store.

All services are configured to restart automatically (restart: unless-stopped), ensuring resilience in a production environment.

Contributing

Please refer to the documentation within each agent's directory for more details on their specific implementation and how to contribute.