llama-index-tools-agentbay 0.1.4

AgentBay tools integration for LlamaIndex - provides browser automation, file operations, and command execution capabilities

LlamaIndex Tools Integration: AgentBay

AgentBay tools integration for LlamaIndex, enabling browser automation, file operations, and command execution.

Installation

pip install llama-index-tools-agentbay llama-index-llms-openai-like

Setup

Set your API keys as environment variables:

export AGENTBAY_API_KEY="your-agentbay-api-key"
export DASHSCOPE_API_KEY="your-dashscope-api-key"  # Optional: For DashScope LLM

Quick Start

Use the context manager for automatic resource cleanup:

import os
import asyncio
from llama_index.llms.openai_like import OpenAILike
from llama_index.core.agent import ReActAgent
from llama_index.tools.agentbay import create_code_tools, AgentBaySessionManager
from contextlib import contextmanager

@contextmanager
def agentbay_tools():
    manager = AgentBaySessionManager(api_key=os.getenv("AGENTBAY_API_KEY"))
    try:
        yield create_code_tools(manager)
    finally:
        manager.cleanup()

async def main():
    llm = OpenAILike(
        model="qwen-plus",
        api_key=os.getenv("DASHSCOPE_API_KEY"),
        api_base="https://dashscope.aliyuncs.com/compatible-mode/v1",
        is_chat_model=True
    )

    with agentbay_tools() as tools:
        # Note: ReActAgent in llama-index-core >= 0.14.0 is workflow-based
        agent = ReActAgent.from_tools(tools, llm=llm, verbose=True)
        response = await agent.achat("Calculate the 10th Fibonacci number using Python")
        print(response.response)

if __name__ == "__main__":
    asyncio.run(main())

Available Tools

• Code Execution: create_code_tools (Recommended) - Run Python/JS code directly.
• Browser Automation: create_browser_tools - Take screenshots, inspect pages.
• File Operations: create_filesystem_tools - Read/Write files.
• Command Execution: create_command_tools - Run shell commands.

RAG Integration

Extract insights from execution results:

from llama_index.tools.agentbay import create_rag_manager

rag = create_rag_manager()
rag.add_execution_result("result content", "task description")
print(rag.query("What was the result?"))

Support

• AgentBay Console
• AgentBay SDK Documentation
• LlamaIndex Documentation

License

MIT