nat.agent.auto_memory_wrapper.register#

Attributes#

logger

Classes#

AutoMemoryAgentConfig

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

Functions#

auto_memory_agent(...)

Build the auto-memory agent that wraps another agent.

Module Contents#

logger#

class AutoMemoryAgentConfig(/, **data: Any)#

Bases: nat.data_models.agent.AgentBaseConfig

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

This agent automatically captures user messages, retrieves relevant context, and stores agent responses without requiring the LLM to invoke memory tools.

Use this when: - You want guaranteed memory capture (not dependent on LLM tool calling) - You need consistent memory operations across all interactions - Your memory backend (Zep, Mem0) is designed for automatic memory management

Use tool-based memory when: - You want the LLM to decide when to access memory - Memory operations should be selective based on context

Example:

functions:
  my_react_agent:
    _type: react_agent
    llm_name: nim_llm
    tool_names: [calculator, web_search]

memory:
  zep_memory:
    _type: nat.plugins.zep_cloud/zep_memory

workflow:
  _type: auto_memory_agent
  inner_agent_name: my_react_agent
  memory_name: zep_memory
  llm_name: nim_llm
  verbose: true

Multi-tenant User Isolation:

User ID is automatically extracted from runtime context (user_manager.get_id()) for proper multi-tenant memory isolation. Set user_manager via SessionManager.session() in production. Defaults to “default_user” for testing/development. See README.md for deployment examples.

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

memory_name: nat.data_models.component_ref.MemoryRef = None#

inner_agent_name: nat.data_models.component_ref.FunctionRef = None#

save_user_messages_to_memory: bool = None#

retrieve_memory_for_every_response: bool = None#

save_ai_messages_to_memory: bool = None#

search_params: dict[str, Any] = None#

add_params: dict[str, Any] = None#

async auto_memory_agent( config: AutoMemoryAgentConfig, builder: nat.builder.builder.Builder, ) → collections.abc.AsyncGenerator[nat.builder.function_info.FunctionInfo, None]#

Build the auto-memory agent that wraps another agent.

The inner agent is retrieved as a Function that receives a ChatRequest with multiple messages (including system messages with memory context). It manages its own internal state (ReActGraphState, etc.) and the wrapper never manipulates that state.