Skip to content

Main Agent

Main agent module for initializing and running the Talk2Scholars application.

This module sets up the hierarchical agent system using LangGraph and integrates various sub-agents for handling different tasks such as semantic scholar, zotero, PDF processing, and paper downloading.

Functions: - get_app: Initializes and returns the LangGraph-based hierarchical agent system.

get_app(uniq_id, llm_model)

Initializes and returns the LangGraph-based hierarchical agent system.

This function constructs the agent workflow by defining nodes for the supervisor and sub-agents. It compiles the graph using StateGraph to enable structured conversational workflows.

Parameters:

Name Type Description Default
thread_id str

A unique session identifier for tracking conversation state.

 required
llm_model BaseChatModel

The language model used for query processing. Defaults to ChatOpenAI(model="gpt-4o-mini", temperature=0).

 required

Returns:

Name Type Description
StateGraph

A compiled LangGraph application that can process user queries.
Example

app = get_app("thread_123") result = app.invoke(initial_state)

Source code in aiagents4pharma/talk2scholars/agents/main_agent.py 
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
def get_app(uniq_id, llm_model: BaseChatModel):
    """
    Initializes and returns the LangGraph-based hierarchical agent system.

    This function constructs the agent workflow by defining nodes for the supervisor
    and sub-agents. It compiles the graph using `StateGraph` to enable structured
    conversational workflows.

    Args:
        thread_id (str): A unique session identifier for tracking conversation state.
        llm_model (BaseChatModel, optional): The language model used for query processing.
            Defaults to `ChatOpenAI(model="gpt-4o-mini", temperature=0)`.

    Returns:
        StateGraph: A compiled LangGraph application that can process user queries.

    Example:
        >>> app = get_app("thread_123")
        >>> result = app.invoke(initial_state)
    """
    # Replace placeholder mini model with a configured ChatOpenAI instance
    if getattr(llm_model, "model_name", None) == "gpt-4o-mini":
        llm_model = ChatOpenAI(
            model="gpt-4o-mini",
            temperature=0,
            model_kwargs={"parallel_tool_calls": False},
        )
    # Load hydra configuration
    logger.log(logging.INFO, "Launching Talk2Scholars with thread_id %s", uniq_id)
    with hydra.initialize(version_base=None, config_path="../configs/"):
        cfg = hydra.compose(
            config_name="config", overrides=["agents/talk2scholars/main_agent=default"]
        )
        cfg = cfg.agents.talk2scholars.main_agent
    logger.log(logging.INFO, "System_prompt of Talk2Scholars: %s", cfg.system_prompt)
    # Create supervisor workflow
    workflow = create_supervisor(
        [
            get_app_s2(uniq_id, llm_model),  # semantic scholar
            get_app_zotero(uniq_id, llm_model),  # zotero
            get_app_paper_download(uniq_id, llm_model),  # pdf
            get_app_pdf(uniq_id, llm_model),  # paper download
        ],
        model=llm_model,
        state_schema=Talk2Scholars,
        # Full history is needed to extract
        # the tool artifacts
        output_mode="full_history",
        # Allow supervisor to resume control and chain multiple sub-agent calls
        add_handoff_back_messages=True,
        prompt=cfg.system_prompt,
    )

    # Compile and run
    app = workflow.compile(checkpointer=MemorySaver(), name="Talk2Scholars_MainAgent")

    return app