neurotrace.core.tools.memory

neurotrace.core.tools.memory

memory_search_tool(memory_orchestrator, tool_name='search_memory', tool_description=None, **kwargs)

Creates a tool that searches memory (vector + graph) and returns fused results.

Parameters:

Name	Type	Description	Default
memory_orchestrator	MemoryOrchestrator	Manages both vector and graph memory.	required
tool_name	str	Name of the tool. Defaults to "search_memory".	'search_memory'
tool_description	str	Description shown to the agent. Loaded from prompt if None.	None
**kwargs		Other Tool configuration options.	{}

Returns:

Name	Type	Description
Tool	Tool	A LangChain tool for searching across memory.

Source code in neurotrace/core/tools/memory.py

def memory_search_tool(
    memory_orchestrator: MemoryOrchestrator,
    tool_name: str = "search_memory",
    tool_description: str = None,
    **kwargs,
) -> Tool:
    """
    Creates a tool that searches memory (vector + graph) and returns fused results.

    Args:
        memory_orchestrator (MemoryOrchestrator): Manages both vector and graph memory.
        tool_name (str): Name of the tool. Defaults to "search_memory".
        tool_description (str): Description shown to the agent. Loaded from prompt if None.
        **kwargs: Other Tool configuration options.

    Returns:
        Tool: A LangChain tool for searching across memory.
    """

    def _search(query: str) -> str:
        """
        Searches both vector and graph memory for relevant info.

        Args:
            query (str): The question or search query.

        Returns:
            str: Combined result from vector and graph memory.
        """

        # Vector memory: semantic search
        vector_results = memory_orchestrator.search_vector_memory(query)
        vector_summary = (
            "\n".join(f"- {doc.content}" for doc in vector_results)
            if vector_results
            else "No relevant vector memory found."
        )

        # Graph memory: entity/triplet reasoning
        graph_result = memory_orchestrator.search_graph_memory(query)
        graph_summary = graph_result if graph_result else "No graph relationships found."

        summarised_memory_context = perform_summarisation(
            llm=memory_orchestrator.llm,
            prompt=PROMPT_SUMMARISE_VECTOR_AND_GRAPH_MEMORY,
            prompt_placeholders={
                "vector_memory": vector_summary,
                "graph_memory": graph_summary,
            },
        )

        # Combine and return
        return (
            "This is the summarised context from both vector and graph memory:\n"
            "--- START OF MEMORY CONTEXT ---\n\n"
            f"{summarised_memory_context}\n\n"
            "--- END OF MEMORY CONTEXT ---"
        )

    return generic_tool_factory(
        func=_search,
        tool_name=tool_name,
        tool_description=tool_description or load_prompt(tool_name),
        **kwargs,
    )

save_memory_tool(memory_orchestrator, tool_name='save_memory', tool_description=None, **kwargs)

Creates a tool that allows the agent to explicitly save important memories to long-term vector memory.

Parameters:

Name	Type	Description	Default
vector_memory_adapter		The adapter to store messages.	required
tool_name	str	Name of the tool. Defaults to "save_memory".	'save_memory'
tool_description	str	Description of the tool. Loads from prompt if None.	None
**kwargs		Additional keyword args for Tool.	{}

Returns:

Name	Type	Description
Tool	Tool	A configured LangChain Tool instance for saving memory.
	Tool	param tool_name:
	Tool	param memory_orchestrator:

Source code in neurotrace/core/tools/memory.py

def save_memory_tool(
    memory_orchestrator: MemoryOrchestrator,
    tool_name: str = "save_memory",
    tool_description: str = None,
    **kwargs,
) -> Tool:
    """
    Creates a tool that allows the agent to explicitly save important memories
    to long-term vector memory.

    Args:
        vector_memory_adapter: The adapter to store messages.
        tool_name (str, optional): Name of the tool. Defaults to "save_memory".
        tool_description (str, optional): Description of the tool. Loads from prompt if None.
        **kwargs: Additional keyword args for Tool.

    Returns:
        Tool: A configured LangChain Tool instance for saving memory.
        :param tool_name:
        :param memory_orchestrator:
    """

    def _save(summary: str) -> str:
        """
        Saves a summary in vector memory using the orchestrator.

        Args:
            summary (str): The summary to save.

        Returns:
            str: Confirmation message indicating the summary was saved.
        """
        message_text, convo_tags, *_ = summary.split("-- tags:") + [None, None]
        if convo_tags:
            convo_tags = convo_tags.strip().split(",")
        else:
            convo_tags = []

        message_text = message_text.strip()

        _save_in_vector_memory(memory_orchestrator, message_text, convo_tags)
        _save_in_graph_memory(memory_orchestrator, message_text, convo_tags)
        return "Memory saved in both vector and graph memory."

    return generic_tool_factory(
        func=_save,
        tool_name=tool_name,
        tool_description=tool_description or load_prompt(tool_name),
        **kwargs,
    )