Mono-Agent Framework

A minimal, file-based agent framework inspired by smolagents and LangGraph, optimized for offline use and context efficiency.

Features

• CLI Scaffolding: Quickly start projects with mono create.
• Markdown Agents: Define agent personas and tools in simple Markdown files.
• Auto-Tool Sync: Python functions in Tools/ are automatically converted to OpenAI JSON schemas.
• Context Efficient: Automatically wipes previous agent context during handoffs to save tokens.
• Persistent Memory: SQLite-based conversation history and Mem0 fact extraction.
• Human-in-the-Loop: Built-in support for human validation and steering.

Installation

Install the framework using your preferred package manager:

# Using uv (Recommended)
uv add mono-agent

# Using pip
pip install mono-agent

Quick Start

Once installed, you can set up a new project in seconds using the CLI:

1. Scaffold your project: This command creates the Agents/ and Tools/ folders, a sample agent, and a main.py entry point. It also helps you configure your LLM provider.

   mono create
   

2. Test your setup: Perform a dynamic dry run to ensure all files and configurations are correct.

   mono test
   

3. Run your agent:

   python main.py
   

Creating a Specialist Agent

1. Create a tool in Tools/my_tool.py:

   def run(query: str) -> str:
       """Does something cool."""
       return f"Result for {query}"
   

2. Generate the Agent file:

   from mono_agent import MonoAgent
   agent = MonoAgent()
   agent.create_agent_file(
       name="MySpecialist",
       instructions="You are a specialist...",
       tool_names=["my_tool"]
   )
   

3. Load and Run:

   agent.load_agent("MySpecialist")
   agent.run("Do the cool thing.")
   

License

MIT License. See LICENSE for details.