augagent 0.1.0

A multi-agent Python framework with intuitive DX, powered by Pydantic.

AugAgent 🤖

AugAgent is a multi-agent framework designed in Python, providing a deeply intuitive Developer Experience (DX) heavily inspired by industry standards like CrewAI and AutoGen, but supercharged with Pydantic for rigid type safety and data validation.

By utilizing strict Pydantic schemas under the hood, AugAgent ensures that Language Models produce predictable, structured outputs, effectively eliminating hallucinated parameters and malformed function arguments before they ever reach your runtime code.

🚀 Why AugAgent?

• Familiar, Intuitive DX: If you've used CrewAI or AutoGen, you'll feel right at home. The AugAgent (Role/Goal/Backstory), AugTask, and AugTeam abstractions are heavily aligned with the latest industry orchestration standards.
• Pydantic Native: LLM configurations, chat schemas, and tool arguments are built on Pydantic v2.
• OpenAI Compatible: Easily plug in any OpenAI-compatible API endpoint (e.g. OpenAI, Azure, vLLM, Ollama) directly into the LLMConfig.
• Automatic Tool Schemas: The @aug_tool decorator dynamically generates strict OpenAI function-calling JSON schemas from standard Python type hints.
• Context Chaining: AugTeam orchestrators automatically chain task outputs as context for subsequent tasks.

---

🛠️ Quick Start

Install AugAgent:

pip install augagent

Below is an example of an orchestrated Research and Writing team using the framework:

import asyncio
from augagent import AugAgent, AugTask, AugTeam, aug_tool, LLMConfig
from pydantic import Field

# 1. Define tools

@aug_tool
def search_web(query: str = Field(description="Search query")) -> str:
    """Search the web for information."""
    print(f"  [Tool Executing] Searching the web for: {query}")
    # Mock search results for demonstration
    if "quantum" in query.lower():
        return "Quantum computing uses qubits. It can solve complex problems faster than classical computers."
    return f"Found generic results for {query}."

@aug_tool
def fetch_article(url: str = Field(description="URL to fetch")) -> str:
    """Fetch the contents of an article."""
    print(f"  [Tool Executing] Fetching article: {url}")
    return "This is the content of the article regarding " + url

# 2. Define agents

llm_config = LLMConfig(model="gpt-4o-mini", temperature=0.2)

researcher = AugAgent(
    name="Researcher",
    role="Senior Technology Researcher",
    goal="Find comprehensive and accurate information on emerging technologies.",
    backstory="You have 10 years of experience researching deep tech.",
    llm_config=llm_config,
    tools=[search_web, fetch_article],
    verbose=True
)

writer = AugAgent(
    name="Writer",
    role="Technical Content Writer",
    goal="Write engaging, accurate articles about technology.",
    backstory="You write clear, accessible articles for a wide audience.",
    llm_config=llm_config,
    verbose=True
)

# 3. Define tasks

research_task = AugTask(
    description="Research the topic: {topic}. Find key facts and recent advancements.",
    expected_output="A bulleted list of key facts about {topic}.",
    agent=researcher
)

writing_task = AugTask(
    description="Write a short, engaging article about {topic} based on the research provided.",
    expected_output="A 2-paragraph article in Markdown format.",
    agent=writer
)

# 4. Assemble the team and kickoff

team = AugTeam(
    agents=[researcher, writer],
    tasks=[research_task, writing_task],
    verbose=True
)

if __name__ == "__main__":
    print(f"=== Kicking off AugTeam ===")
    results = team.kickoff(inputs={"topic": "Quantum Computing"})
    
    print("\n=== Final Results ===")
    for idx, result in enumerate(results):
        print(f"\nTask {idx + 1} ({result.agent_name}):")
        print("-" * 40)
        print(result.output)

📚 Documentation

• Architecture & Design Choices (WHY)
• Antigravity Ledger (HOW)