agentdev 0.5.0

AgentDev provides basic components that used on Bailian.platform with unified API

AgentDev

Aliyun Bailian Website

Overview

AgentDev is a component-based framework for building AI-powered applications. Key features include:

• Modular component architecture
• Integration with Dashscope services
• LangGraph compatibility
• Model Context Protocol (MCP) support
• Enterprise level tracing and metrics
• Extensible error handling system

Design Principles

1. Components have single responsibility
2. Low replacement cost with high reusability component
3. Quick testing and deployment for individual components
4. Clear dependency relationships through well-defined input/output parameters
5. Compatibility with other open-source solutions
6. Extensible base classes supporting logging and tracing

Key Components

1. Bailian Search component: Enterprise level web search based on bailian official search ability.
2. Bailian RAG component: Enterprise level RAG which retrieve relevant information from Bailian maintain by user.
3. Bailian Intention component: Enterprise level Intention Center which classify user's intention with different label.
4. Aliyun Tracing component: Enterprise level tracing that manage user tracing data by aliyun
5. OpenAI and Bailian compatible LLM message schemas
6. OpenAI compatible LLM component
7. FastAPI server for SSE protoc
8. Other key components as base capabilities to support Bailian Chat Services

Installation

pip install agentdev

Quick Start

Basic llm service

Create a python file named main.py with the following code. Notice that the DashscopeChatRequest is the original Dashscope chat request format, if you want to use openai compatible request, you can use OpenAIChatRequest or BailianChatRequest instead.

from agentdev.models.llm import BaseLLM
from agentdev.schemas.llm_schemas import DashscopeChatRequest
from agentdev.server.fastapi_server import FastApiServer


async def general_arun(request: DashscopeChatRequest):
    """Test arun method"""
    llm = BaseLLM()
    try:
        async for chunk in await llm.arun(model=request.model, messages=request.input.messages,
                                          parameters=request.parameters):
            yield chunk
    except Exception as e:
        import traceback
        print(f'error {e} with traceback {traceback.format_exc()}')


server = FastApiServer(
    func=general_arun,
    endpoint_path='/api/v1/chat/completions',
    request_model=DashscopeChatRequest,
)

if __name__ == '__main__':
    server.run()

Then run the python file with your DASHSCOPE_API_KEY

export DASHSCOPE_API_KEY=your_api_key
python main.py

then a local llm service will run on your local machine at http://127.0.0.1:8000/api/v1/chat/completions

query the service with curl

curl --location 'http://127.0.0.1:8000/api/v1/chat/completions' \
--header 'Content-Type: application/json' \
--data '{
    "model": "qwen-max",
    "input":{
        "messages":[{"role": "user", "content":"今天天气如何"}]
    },
    "parameters": {
    }
}'

the response should be like this

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"","role":"assistant"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"我"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"目前"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"无法直接获取实时天气"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"信息。您可以通过查询网络"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"、使用天气应用查看"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"，或者告诉我您"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"所在的城市和地区，我可以提供"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"一些一般性的天气信息和"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"建议。如果需要准确"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"的天气情况，请参考"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":"当地的气象预报。"},"index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

data:{"id":"chatcmpl-2b0cb0d9-82fa-9c2c-8364-c100111d84a3","choices":[{"delta":{"content":""},"finish_reason":"stop","index":0}],"created":1742953771,"model":"qwen-max","object":"chat.completion.chunk"}

Basic component create

import asyncio

from agentdev.base.component import Component
from pydantic import BaseModel, Field
from typing import Any

# define the input format
class SearchInput(BaseModel):
    """
    Search Input.
    """
    query: str = Field(..., title='Query')
    context: dict


# define the output format
class SearchOutput(BaseModel):
    """
    Search Output.
    """
    results: list[str]
    context: dict


# define the component runtime method
class SearchComponent(Component[SearchInput, SearchOutput]):
    """
    Search Component.
    """

    async def arun(self, args: SearchInput, **kwargs: Any) -> SearchOutput:
        """
        Run asynchronously.
        """
        if not isinstance(args, SearchInput):
            raise TypeError(
                'Argument must be an instance of SearchInput or its subclass')

        await asyncio.sleep(1)  # 模拟异步操作
        return SearchOutput(results=['result1', 'result2'],
                            context=args.context)


# instantiate the component
search_component = SearchComponent(
    name='Search Component',
    description='Search Component For Example',
    config={})
search_input = SearchInput(query='query', context={})


# run
async def main():
    search_output = await search_component.arun(search_input)
    print(search_output)
    print(search_component.function_schema)


asyncio.run(main())

Requirements

1. Quick deployment similar to assistant serving
2. Standardized output like Marvin
3. Agent evaluation capabilities

Solutions Implemented

1. Single-responsibility component interfaces with clear input/output definitions
2. Extensible base classes with optional logging/tracing
3. Mixins for integrating with open-source ecosystems

Development

Testing

Run tests using pytest:

pytest test/

Contributing

Please follow these guidelines:

• Write unit tests for new components
• Maintain consistent documentation
• Follow existing code style and patterns