Functions
tuple[dict, dict]	_extract_plan (APlannerState state)
	_plan_value (APlannerState state, str key, default=None)
	invoke_db_helper (APlannerState state)
	invoke_web_helper (APlannerState state)
	route_from_planning (APlannerState state)
APlannerState	answer_node (APlannerState state)

Variables
	PARENT_DIR = os.path.abspath(os.path.join(os.path.dirname(__file__), '..'))
	CONFIG_PATH = os.path.join(PARENT_DIR, "config.json")
	CONFIG = json.load(f)
	LOOP_CONFIG = CONFIG["loop_limits"]
	graph_builder = StateGraph(APlannerState)
	a_planner_node
	defer
	a_chat_graph = graph_builder.compile()

Detailed Description

Copyright 2026 Luca Silver

Advisor-facing chat agent that routes advisor queries through planning and helper graphs.

Functions:
- `invoke_db_helper`: Invokes the database helper graph and merges results back to planner state.
- `invoke_web_helper`: Invokes the web search helper graph and merges results back to planner state.
- `route_from_planning`: Routes the planner's decision to appropriate helper graphs or answer node. Enforces a maximum number of loops before forcing an answer.
- `answer_node`: Appends the final planner response to the conversation message history.

Graph Structure:
- START -> planning
- planning -> [invoke_db_helper | invoke_web_helper | answer_node] (conditional) (cam be parallel)
- invoke_db_helper -> planning (feedback loop) (deferred)
- invoke_web_helper -> planning (feedback loop) (deferred)
- answer_node -> END

Exports:
- `a_chat_graph`: Compiled LangGraph advisor chat agent.

Loop limits are defined in config.json.

Function Documentation

◆ _extract_plan()

tuple[dict, dict] _extract_plan ( APlannerState state )

protected

Safely extracts planner output and nested content dict from state.

Some model/tooling paths may produce plan payloads where "content" is a
string or missing entirely. This helper normalizes both objects so routing
code can perform key lookups without type errors.

◆ _plan_value()

_plan_value	(	APlannerState	state,
		str	key,
			default = None )

protected

Reads a planner field from either top-level plan or nested content.

◆ answer_node()

APlannerState answer_node ( APlannerState state )

Appends the final planner answer to the conversation state.

Args:
    state (APlannerState): Current planner state containing the final answer in
        plan["answer"].

Returns:
    APlannerState: Updated state with the final AI answer appended to messages.

◆ invoke_db_helper()

invoke_db_helper ( APlannerState state )

Invokes the advisor database helper graph and merges its result back into state.

The advisor planner may request database information about a student or course.
This node forwards that request to the database helper graph, then appends the
returned query/result pair to the accumulated db_info list.

Args:
    state (APlannerState): Current planner state containing the selected database
        request in plan["info_needed_db"].

Returns:
    dict: Partial state update with the updated "db_info" list.

◆ invoke_web_helper()

invoke_web_helper ( APlannerState state )

Invokes the web helper graph and merges its result back into state.

The advisor planner may request web information to support an answer. This node
forwards that request to the web helper graph, then appends the returned
query/result pair to the accumulated web_info list.

Args:
    state (APlannerState): Current planner state containing the selected web
        request in plan["info_needed_web"].

Returns:
    dict: Partial state update with the updated "web_info" list.

◆ route_from_planning()

route_from_planning ( APlannerState state )

Chooses which helper graphs to run after planning.

The planner can request database lookups or web lookups, or provide a final
answer directly. This router converts the plan into one or more graph sends.
When both database and web information are requested, they are dispatched in
parallel. If no tools are needed, control routes directly to the answer node.

Args:
    state (APlannerState): Current planner state containing the loop counter and
        the structured plan produced by the planning node.

Returns:
    list[Send]: One or more graph sends describing the next execution branch.

Variable Documentation

◆ a_chat_graph

lg_agent.a_chat_graph.a_chat_graph = graph_builder.compile()

◆ a_planner_node

lg_agent.a_chat_graph.a_planner_node

◆ CONFIG

lg_agent.a_chat_graph.CONFIG = json.load(f)

◆ CONFIG_PATH

lg_agent.a_chat_graph.CONFIG_PATH = os.path.join(PARENT_DIR, "config.json")

◆ defer

lg_agent.a_chat_graph.defer

◆ graph_builder

lg_agent.a_chat_graph.graph_builder = StateGraph(APlannerState)

◆ LOOP_CONFIG

lg_agent.a_chat_graph.LOOP_CONFIG = CONFIG["loop_limits"]

◆ PARENT_DIR

lg_agent.a_chat_graph.PARENT_DIR = os.path.abspath(os.path.join(os.path.dirname(__file__), '..'))

Functions

Variables