nat.agent.auto_memory_wrapper.agent#

Attributes#

logger

Classes#

AutoMemoryWrapperGraph

Wraps any NAT agent to add automatic memory capture and retrieval.

Module Contents#

logger#

class AutoMemoryWrapperGraph( inner_agent_fn: collections.abc.Callable[[nat.data_models.api_server.ChatRequest], collections.abc.Awaitable[Any]], memory_editor: nat.memory.interfaces.MemoryEditor, save_user_messages: bool = True, retrieve_memory: bool = True, save_ai_responses: bool = True, search_params: dict[str, Any] | None = None, add_params: dict[str, Any] | None = None, )#

Wraps any NAT agent to add automatic memory capture and retrieval.

The wrapper treats the inner agent as a black-box function that receives a ChatRequest with multiple messages (including system messages with memory context). The inner agent manages its own internal state (ReActGraphState, ReWOOGraphState, etc.) - the wrapper never sees it.

inner_agent_fn#

memory_editor#

save_user_messages = True#

retrieve_memory = True#

save_ai_responses = True#

search_params#

add_params#

_context#

_get_user_id_from_context() → str#

Extract user_id from runtime context.

Priority order: 1. user_manager.get_id() - For authenticated sessions (set via SessionManager.session()) 2. X-User-ID HTTP header - For testing/simple auth without middleware 3. “default_user” - Fallback for development/testing without authentication

Returns:: str: The user ID for memory operations

get_wrapper_node_count() → int#

Calculate the number of wrapper nodes that will be added to the graph.

This is used to determine the recursion_limit overhead that the wrapper adds on top of the inner agent’s recursion needs.

Returns:: int: Number of nodes in the wrapper graph (1-4 depending on configuration)

static _langchain_message_to_nat_message( lc_message: langchain_core.messages.BaseMessage, ) → nat.data_models.api_server.Message#

Convert LangChain message to NAT Message format.

This is necessary to construct a proper ChatRequest with all messages (including system messages with memory context) to pass to the inner agent.

async capture_user_message_node( state: nat.agent.auto_memory_wrapper.state.AutoMemoryWrapperState, ) → nat.agent.auto_memory_wrapper.state.AutoMemoryWrapperState#: Captures user message to memory thread

async memory_retrieve_node( state: nat.agent.auto_memory_wrapper.state.AutoMemoryWrapperState, ) → nat.agent.auto_memory_wrapper.state.AutoMemoryWrapperState#: Retrieves relevant memory from memory store

async inner_agent_node( state: nat.agent.auto_memory_wrapper.state.AutoMemoryWrapperState, ) → nat.agent.auto_memory_wrapper.state.AutoMemoryWrapperState#

Calls the inner agent with a ChatRequest containing all messages.

The inner agent receives a ChatRequest with multiple messages (including system messages with memory context), processes them using its own internal state (ReActGraphState, ReWOOGraphState, etc.), and returns a ChatResponse.

async capture_ai_response_node( state: nat.agent.auto_memory_wrapper.state.AutoMemoryWrapperState, ) → nat.agent.auto_memory_wrapper.state.AutoMemoryWrapperState#: Captures agent response to memory

build_graph() → langgraph.graph.state.CompiledStateGraph#: Wraps inner agent with memory nodes