Agent Zero Lite

A lightweight, cross-platform implementation of Agent Zero that maintains core functionality while reducing complexity and dependencies.

Features

✅ Full LiteLLM Support - 100+ AI providers (OpenAI, Anthropic, Google, local models, etc.)
✅ Web UI - Complete interface at localhost:50001
✅ Vector Memory - FAISS-based persistent memory
✅ Document RAG - PDF, text, and document processing
✅ Multi-Agent - Superior/subordinate agent hierarchy
✅ MCP Client - Model Context Protocol integration
✅ Local Execution - Python, Node.js, and terminal
✅ Tunneling - Remote access support
✅ File Management - Work directory browser

Removed from Full Version

❌ Browser automation (Playwright)
❌ Docker/SSH execution
❌ Speech processing (STT/TTS)
❌ Task scheduling
❌ Backup/restore system
❌ Web search tools

Quick Start

1. Install (CPU-only by default):

pip install agent-zero-lite

• Optional extras:
  • CPU ML helpers (additional ONNX/Transformers utilities):

    pip install "agent-zero-lite[cpu]"
    

  • Transformers stack (CPU) and ONNX runtime (sentence-transformers included by default):

    pip install "agent-zero-lite[ml]"
    

  • Audio transcription (Whisper, CPU):

    pip install "agent-zero-lite[audio]"
    

  • GPU stack (advanced; choose your CUDA build of torch separately if needed):

    pip install "agent-zero-lite[gpu]"
    # For PyTorch CUDA builds, see: https://pytorch.org/get-started/locally/
    

2. Configure environment:

cp .env.example .env
# Edit .env with your API keys

3. Start the Web UI:

python run_ui.py

4. Open browser:

   http://localhost:50001
   

Configuration

Minimal Setup

Set at least one LLM provider in .env:

# OpenAI
OPENAI_API_KEY=sk-...

# Or Anthropic
ANTHROPIC_API_KEY=sk-ant-...

# Or local Ollama
CHAT_MODEL_PROVIDER=ollama
CHAT_MODEL_NAME=llama3.1:8b
OLLAMA_API_BASE=http://localhost:11434

Full Configuration

See .env.example for all available options including:

• All 100+ LiteLLM providers
• Model configurations
• Rate limiting settings
• MCP server integration
• Memory and knowledge settings

Supported Models

Agent Zero Lite supports all LiteLLM providers:

Commercial APIs

• OpenAI: GPT-4o, GPT-4, GPT-3.5, etc.
• Anthropic: Claude 3.5 Sonnet, Claude 3 Opus, etc.
• Google: Gemini 1.5 Pro, Gemini 1.5 Flash, etc.
• Groq: Llama 3.1, Mixtral, etc. (fast inference)
• Together AI: Llama, Mistral, etc.
• Mistral AI: Mistral Large, Mistral 7B, etc.
• Cohere: Command R+, Command Light, etc.

Local Models

• Ollama: Any local model (llama3.1, mistral, etc.)
• LM Studio: Local model server
• Text Generation WebUI: Local inference
• VLLM: High-performance inference server

Enterprise

• Azure OpenAI: Enterprise GPT models
• AWS Bedrock: Claude, Titan, etc.
• Google Vertex AI: Enterprise Gemini
• Hugging Face: Hosted models

Usage Examples

Basic Chat

from agent import AgentContext
import initialize

# Initialize agent
config = initialize.initialize_agent()
context = AgentContext(config)

# Send message
response = context.communicate("Hello, what can you help me with?")

Code Execution

The agent can execute Python, Node.js, and terminal commands:

User: "Create a Python script that calculates fibonacci numbers"
Agent: Uses code_execution tool to write and run Python code

Document Processing

User: "Analyze this PDF document and summarize the key points"
Agent: Uses document_query tool to process and analyze documents

Multi-Agent Collaboration

User: "Create a complex analysis using multiple specialized agents"
Agent: Uses call_subordinate to delegate tasks to specialized sub-agents

Architecture

Agent Zero Lite maintains the core Agent Zero architecture:

• Agent Loop: Reason → Tool Use → Response cycle
• Tool System: Extensible plugin architecture
• Memory: FAISS vector database for persistent memory
• Extensions: Hook-based system for customization
• Prompts: Template-based prompt management

Development

Adding Tools

Create new tools in python/tools/:

from python.helpers.tool import Tool, Response

class MyTool(Tool):
    async def execute(self, **kwargs):
        # Tool logic here
        return Response(message="result", break_loop=False)

Adding Extensions

Create extensions in python/extensions/:

from python.helpers.extension import Extension

class MyExtension(Extension):
    async def execute(self, **kwargs):
        # Extension logic here
        pass

Troubleshooting

Common Issues

1. Model not responding: Check API keys in .env
2. Port in use: Change PORT in .env
3. Memory issues: Reduce context length settings
4. Missing dependencies: Run pip install -r requirements.txt

Debugging

Enable debug logging by setting:

LITELLM_LOG=DEBUG

Migration

From Full Agent Zero

1. Copy .env settings
2. Copy memory/ and knowledge/ folders
3. Copy work_dir/ contents
4. Remove Docker/SSH configurations

To Full Agent Zero

1. Install additional dependencies
2. Add Docker/SSH configurations
3. No data migration needed

Performance

Agent Zero Lite is optimized for:

• Startup: ~3 seconds vs 15+ seconds
• Memory: ~200MB vs 1GB+ RAM usage
• Dependencies: ~30 packages vs 45+ packages
• Installation: <2 minutes vs 10+ minutes

License

Same as Agent Zero - check the original repository for license terms.

Support

For issues and questions:

1. Check this README
2. Review .env.example configuration
3. See the original Agent Zero documentation
4. Report issues to the Agent Zero repository