Agent Development Guide for ChainStream

Chainstream Agent Development Module Overview

Stream Module

• Description: The Stream class forms the core of data streams, where each data stream is an instance of Stream identified by a stream_id. ChainStream enables data monitoring and processing by attaching listener functions to these data streams.
• API:
• chainstream.get_stream(stream_id): Retrieves a Stream object based on the provided stream_id, typically used when constructing an Agent instance to obtain input and output streams.
• chainstream.create_stream(stream_id): Creates a new data stream and returns a Stream instance identified by stream_id.
• chainstream.stream.Stream.for_each(agent, listener_func): Attaches a listener function to the Stream instance.
• chainstream.stream.Stream.unregister_all(agent): Unregisters all listener functions attached to the data stream.
• chainstream.stream.Stream.add_item(data): Pushes data into the data stream.

Agent Module

• Description: One or more agents need to be created to accomplish user-specified tasks. Agent instances should inherit from the chainstream.agent.Agent class, passing an agent_id identifier to the superclass, and implement __init__, start(), and stop() methods to handle data stream monitoring, processing, and result output.
• API:
• __init__(agent_id): Initializes a new Agent object for task completion, where agent_id is a mandatory parameter used to identify the agent within the system. Initialization of resources and data streams can be done within this method.
• start(): Defines the listener function for processing data streams and binds it to the respective data stream.
• stop(): Unregisters all listener functions attached by the agent to the data stream upon completion.

BufferContext Module

• Description: If data storage is required after processing, the BufferContext module can be used to create a data container—a queue where data can only be added at the tail and retrieved from the head.
• API:
• chainstream.context.BufferContext(): Instantiates a new BufferContext object to create a data container.
• chainstream.context.BufferContext.add(data): Adds data to the tail of the data container.
• chainstream.context.BufferContext.get(): Retrieves data from the head of the data container.

LLM Module

• Description: The LLM (Language Learning Model) module integrates various models capable of handling different types of input data such as text, images, and audio. Models respond based on input processing requirements and provided data.
• API:
• chainstream.llm.get_model(type): Instantiates an LLM object to obtain a model for processing data.
• chainstream.llm.make_prompt(query, data): Converts processing requirements and input data into a format accepted by the model.
• chainstream.llm.query(prompt): Sends input prompt to the model and returns the model's response.

Agent Development Guide

Agent_id

In ChainStream, Agent_id is a string used to uniquely identify an agent instance.

When creating an agent, it's essential to specify a unique Agent_id to identify and manage different agents within the system.

• Agent_id can be any string that adheres to naming conventions, such as test_agent, arxiv_processor, etc.
• To avoid conflicts, descriptive names are recommended, and special characters or spaces should be avoided.

API Usage

• When creating a new agent, inherit from the chainstream.agent.Agent class and implement its methods.
• In the __init__ method, call the superclass constructor and initialize resources and data streams.
• In the start method, define the listener function for processing data streams and bind it to the respective data stream.
• Apart from the listener function, typical tasks include data preprocessing, parsing, model querying, response handling, and format conversion.
• In the stop method, unregister all listener functions attached by the agent to the data stream.

Naming Conventions

Follow Python naming conventions for modules, classes, functions, and variables.

• Use lowercase letters and underscores _ to separate words for module and file names.
• Use CamelCase for class names.
• Use lowercase letters and underscores _ for function and variable names.

Example Agent

import chainstream as cs
from chainstream.llm import get_model

class TestAgent(cs.agent.Agent):
    def __init__(self):
        super().__init__("test_arxiv_agent")
        self.input_stream = cs.get_stream("all_arxiv")
        self.output_stream = cs.get_stream("cs_arxiv")
        self.llm = get_model(["text"])

    def start(self):
        def process_paper(paper):
            paper_content = paper["abstract"]
            prompt = "Is this abstract related to edge LLM agent? Say 'yes' or 'no'."
            prompt = [{"role": "user", "content": prompt+paper_content}]
            response = self.llm.query(prompt)
            print(response)
            if response == 'Yes':
                print(paper)
                self.output_stream.add_item(paper)

        self.input_stream.for_each(self, process_paper)

    def stop(self):
        self.input_stream.unregister_all(self)