dragen 0.1.0

CodeAct-style AI agent framework for Python

dragen

CodeAct-style AI agent framework for Python.

Installation

pip install dragen

Quick Start

from dragen import Agent

# Create an agent
agent = Agent("gpt-4o")

# Register a tool
def search(args):
    query = args[0]
    return {"results": [f"Found: {query}"]}

agent.register_function("search", search)

# Run a task
result = agent.run("Search for Python tutorials")
print(result)

Features

• CodeAct pattern: LLM writes Python code to accomplish tasks
• Secure sandbox: Code runs in a sandboxed environment
• Tool registration: Expose Python functions as tools for the agent
• Multi-agent support: Share data between agents via Context

Configuration

from dragen import Agent

agent = Agent(
    "gpt-4o",
    max_iterations=10,
    temperature=0.7,
    max_tokens=4096,
    system="You are a helpful assistant"
)

Tool Registration

Simple function

def add(args):
    return args[0] + args[1]

agent.register_function("add", add)

With type information

from dragen import ToolInfo

info = ToolInfo("search", "Search the web") \
    .arg_required("query", "str", "The search query") \
    .returns("dict")

def search(args):
    query = args[0]
    return {"results": ["result1", "result2"]}

agent.register_tool(info, search)

Multi-Agent with Context

from dragen import Agent, Context

ctx = Context()

# Planner writes output to context
planner = Agent("gpt-4o")
planner.to_context(ctx, "plan")
planner.run("Create a research plan about AI")

# Executor reads from context
executor = Agent("gpt-4o")
executor.from_context(ctx, "plan")
executor.run("Execute the research plan")

Environment Variables

Set your API key based on the model provider:

• OPENAI_API_KEY for OpenAI models (gpt-4o, etc.)
• ANTHROPIC_API_KEY for Anthropic models (claude-*, etc.)
• GROQ_API_KEY for Groq models

License

Apache-2.0