Uso de API a corto plazo de memoria de agente con LangGraph
Las aplicaciones LangGraph a menudo necesitan preservar el contexto de trabajo reciente sin devolver la conversación completa al modelo en cada turno.
Agent Memory expone dos ayudantes diferentes a corto plazo para este problema:
get_summary()devuelve un objetoOracleSummarycuyo valorcontentcomprime la transcripción del thread. Se prefiere cuando la compactación solo necesita compresión de transcripción.get_context_card()devuelve un objetoOracleContextCardcuyo valorcontentes un bloque de contexto listo para la petición de datos con resumen de thread, temas de recuperación, registros duraderos relevantes y mensajes raw recientes. Preferir cuando la compactación debe mantener el contexto consciente de la recuperación para el turno actual.
En este artículo, utilizará el middleware de LangGraph en torno a un agente predefinido para que la memoria del agente pueda mantener las giros automáticamente e inyectar una tarjeta de contexto de Oracle cuando la petición de datos en ejecución sea demasiado grande. El middleware compacta la petición de datos después de que pasa un umbral configurado. En este ejemplo, se selecciona get_context_card() porque la compactación debe conservar el contexto que tiene en cuenta la recuperación, no solo una recapitulación de transcripción.
Advertencia: los resúmenes, las tarjetas de contexto, los registros recuperados y las memorias extraídas automáticamente son texto derivado del modelo o recuperado y se deben tratar como no de confianza. Cuando la extracción o el resumen automáticos están activados, el SDK también puede reutilizar ese texto en peticiones de datos posteriores, como extracción de memoria, resumen, tarjeta de contexto o peticiones de datos de agente, antes de que la aplicación tenga la oportunidad de revisar el valor intermedio específico. Revise las salidas que consume la aplicación, evite permitir que el texto derivado de la memoria autorice acciones con privilegios y utilice extract_memories=False o escrituras de memoria explícitas cuando el flujo de trabajo requiera revisión antes de que el texto derivado pueda influir en la extracción posterior o la construcción del contexto.
En este artículo, aprenderá a:
- Configurar la memoria del agente con un modelo
Embedder, un LLM de memoria de Oracle y un modelo LangGraphChatOpenAI - encapsular un agente LangGraph predefinido con middleware que persiste en nuevas vueltas e inyecta la salida
get_context_card().contentcuando aumenta la presión del token y se inicia la compactación de la petición de datos - la respuesta se convierte más tarde con contexto a corto plazo del thread Agent Memory en lugar de volver a enviar la transcripción completa
Consejo: para la configuración de paquetes, consulte Introducción a la memoria del agente. Si necesita una instancia local de Oracle AI Database para este ejemplo, consulte Run Oracle AI Database Locally.
Configurar la memoria del agente y los modelos LangGraph
Cree un cliente de memoria de agente con una conexión o pool de Oracle DB, configure un Embedder para la búsqueda vectorial, proporcione un LLM de memoria de Oracle para la resolución de tarjetas de contexto y utilice ChatOpenAI para el agente LangGraph.
from typing import Any
from langchain.agents import create_agent
from langchain.agents.middleware import AgentMiddleware
from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, RemoveMessage
from langchain_core.messages.utils import count_tokens_approximately
from langchain_openai import ChatOpenAI
from langgraph.graph.message import REMOVE_ALL_MESSAGES
from langgraph.runtime import Runtime
from oracleagentmemory.core.embedders.embedder import Embedder
from oracleagentmemory.core.llms.llm import Llm
from oracleagentmemory.core.oracleagentmemory import OracleAgentMemory
embedder = Embedder(
model="YOUR_EMBEDDING_MODEL",
api_base="YOUR_EMBEDDING_BASE_URL",
api_key="YOUR_EMBEDDING_API_KEY",
)
memory_llm = Llm(
model="YOUR_MEMORY_LLM_MODEL",
api_base="YOUR_MEMORY_LLM_BASE_URL",
api_key="YOUR_MEMORY_LLM_API_KEY",
temperature=0,
)
langgraph_llm = ChatOpenAI(
model="YOUR_CHAT_MODEL",
base_url="YOUR_CHAT_BASE_URL",
api_key="YOUR_CHAT_API_KEY",
temperature=0,
)
db_pool = ... #an oracledb connection or connection pool
agent_memory = OracleAgentMemory(
connection=db_pool,
embedder=embedder,
llm=memory_llm,
)
thread_id = "langgraph_short_term_demo"
user_id = "user_123"
agent_id = "assistant_456"Configuración de Middleware y un Agente Predefinido
El middleware persiste como nuevo usuario y el asistente se convierte en memoria del agente. Una vez que la petición de datos en ejecución cruza un umbral de token, compacta el estado sustituyendo la lista de mensajes completa por un mensaje memory_context_card sintético más una pequeña cola de los últimos giros raw. Esto mantiene el estado de LangGraph compacto y, al mismo tiempo, proporciona un contexto a corto plazo compatible con la recuperación de agentes incorporada.
Este artículo utiliza la compactación basada en token, pero puede adaptar el mismo patrón a otras políticas, como compactar cada pocos giros o después de un disparador específico de la aplicación. Si implementa la compactación sólo de transcripción, llame a summary = thread.get_summary(...) y lea summary.content; no trate get_summary() como una lista de mensajes.
def _message_text(message: BaseMessage | Any) -> str:
content = getattr(message, "content", "")
if isinstance(content, str):
return content
return str(content)
def _is_context_card_message(message: BaseMessage) -> bool:
return isinstance(message, HumanMessage) and (
getattr(message, "name", None) == "memory_context_card"
)
class OracleShortTermMemoryMiddleware(AgentMiddleware):
"""Persist LangGraph turns and compact prompts with an OracleAgentMemory context card.
Notes
-----
- ``before_model()`` receives the current LangGraph message state for this turn.
After compaction, that state already includes the synthetic ``memory_context_card``
message returned by a previous ``before_model()`` call.
- The middleware strips that synthetic message back out before persisting or
measuring token usage so OracleAgentMemory only stores real user/assistant turns
and the compaction threshold is based on the organic conversation.
- When compaction triggers, the middleware replaces the message history with one
context-card message plus the most recent raw turns. On the next turn, that
same injected message is seen again and filtered out before recomputing the
next compacted prompt.
"""
def __init__(
self,
memory: OracleAgentMemory,
thread_id: str,
user_id: str,
agent_id: str,
compaction_token_trigger: int,
kept_message_count: int,
) -> None:
self._thread = memory.create_thread(
thread_id=thread_id,
user_id=user_id,
agent_id=agent_id,
context_summary_update_frequency=4,
)
self._compaction_token_trigger = int(compaction_token_trigger)
self._kept_message_count = int(kept_message_count)
self._persisted_message_ids: set[str] = set()
def before_model(
self,
state: dict[str, Any],
runtime: Runtime[Any],
) -> dict[str, Any] | None:
del runtime
messages = list(state["messages"])
#^ This will contain the context card message once the compaction occurs
raw_messages = [message for message in messages if not _is_context_card_message(message)]
self._persist_new_messages(raw_messages)
#we exclude the context card from the token counting
if count_tokens_approximately(raw_messages) < self._compaction_token_trigger:
return None
context_card = self._thread.get_context_card().content
if not context_card:
context_card = "<context_card>\n No relevant short-term context yet.\n</context_card>"
return {
"messages": [
RemoveMessage(id=REMOVE_ALL_MESSAGES), #Clear existing message state.
HumanMessage(content=context_card, name="memory_context_card"),
*raw_messages[-self._kept_message_count :],
]
}
def _persist_new_messages(self, messages: list[BaseMessage]) -> None:
persisted: list[dict[str, str]] = []
for message in messages:
#Persist only the conversational roles that map directly to short-
#term memory turns. Tool/system/synthetic messages are skipped here.
role = (
"user"
if isinstance(message, HumanMessage)
else "assistant" if isinstance(message, AIMessage) else None
)
if role is None:
continue
content = _message_text(message).strip()
if not content:
continue
#LangGraph messages usually have stable IDs. When they do not, fall back
#to a content-derived key so the same turn is not persisted repeatedly if
#the caller reuses the returned message list across later invocations.
message_id = str(getattr(message, "id", "") or f"{role}:{hash(content)}")
if message_id in self._persisted_message_ids:
continue
#Track what this middleware instance has already written so each real turn
#is added to Oracle once even though later turns may still carry the same
#messages in the LangGraph state.
self._persisted_message_ids.add(message_id)
persisted.append({"role": role, "content": content})
if persisted:
self._thread.add_messages(persisted)
short_term_middleware = OracleShortTermMemoryMiddleware(
memory=agent_memory,
thread_id=thread_id,
user_id=user_id,
agent_id=agent_id,
compaction_token_trigger=120,
kept_message_count=3,
)
agent = create_agent(
model=langgraph_llm,
tools=[],
middleware=[short_term_middleware],
)Respuesta posteriormente con contexto inyectado en middleware
El usuario Append recurre a la lista de mensajes en ejecución del agente predefinido y permite que el middleware decida cuándo inyectar una tarjeta de contexto. Para el momento en que llegue el turno posterior, el agente puede responder desde un estado compacto que aún contenga contexto a corto plazo de memoria del agente. El ejemplo imprime la tarjeta de contexto inyectada e incluye una muestra recortada para que pueda inspeccionar qué compactación se insertó en la petición de datos sin volcar el bloque completo en línea.
messages: list[BaseMessage] = []
def print_current_context_card(messages: list[BaseMessage]) -> None:
for message in messages:
if _is_context_card_message(message):
print(_message_text(message))
return
print("<context_card>\n No injected context card yet.\n</context_card>")
def run_turn(user_text: str) -> str:
messages.append(HumanMessage(content=user_text))
result = agent.invoke({"messages": messages})
messages[:] = list(result["messages"])
assistant_message = next(
message for message in reversed(messages) if isinstance(message, AIMessage)
)
return _message_text(assistant_message)
run_turn(
"I'm Maya. I'm migrating our nightly invoice reconciliation workflow "
"from cron jobs to LangGraph."
)
run_turn("The failing step right now is ledger enrichment after reconciliation.")
final_answer = run_turn(
"What workflow am I migrating, which step is failing, and who am I?"
)
print_current_context_card(messages)
#<context_card>
#<topics>
#<topic>invoice reconciliation migration</topic>
#<topic>ledger enrichment failure</topic>
#...
#</topics>
#<summary>
#Maya is migrating the nightly invoice reconciliation workflow from cron jobs
#to LangGraph. The failing step is ledger enrichment after reconciliation.
#</summary>
#...
#</context_card>
print(final_answer)
#You're Maya, migrating your nightly invoice reconciliation workflow from cron jobs
#to LangGraph, and the ledger-enrichment step after reconciliation is currently failing.Conclusión
En este artículo, ha aprendido a distinguir get_summary().content de get_context_card().content, a configurar el contexto a corto plazo de la memoria del agente en torno a un agente LangGraph predefinido y a permitir que el middleware compacte la petición de datos con una tarjeta de contexto cuando la conversación sea demasiado grande para mantenerla literal.
Consejo: después de haber aprendido a agregar contexto de thread a corto plazo a un flujo de LangGraph, ahora puede continuar con Uso de memoria de agente con LangGraph.
Código Completo
#Copyright © 2026 Oracle and/or its affiliates.
#This software is under the Apache License 2.0
#(LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0) or Universal Permissive License
#(UPL) 1.0 (LICENSE-UPL or https://oss.oracle.com/licenses/upl), at your option.
#Oracle Agent Memory Code Example - LangGraph Short-Term Memory
#--------------------------------------------------------------
##Configure Oracle Agent Memory and LangGraph models for short term context
from typing import Any
from langchain.agents import create_agent
from langchain.agents.middleware import AgentMiddleware
from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, RemoveMessage
from langchain_core.messages.utils import count_tokens_approximately
from langchain_openai import ChatOpenAI
from langgraph.graph.message import REMOVE_ALL_MESSAGES
from langgraph.runtime import Runtime
from oracleagentmemory.core.embedders.embedder import Embedder
from oracleagentmemory.core.llms.llm import Llm
from oracleagentmemory.core.oracleagentmemory import OracleAgentMemory
embedder = Embedder(
model="YOUR_EMBEDDING_MODEL",
api_base="YOUR_EMBEDDING_BASE_URL",
api_key="YOUR_EMBEDDING_API_KEY",
)
memory_llm = Llm(
model="YOUR_MEMORY_LLM_MODEL",
api_base="YOUR_MEMORY_LLM_BASE_URL",
api_key="YOUR_MEMORY_LLM_API_KEY",
temperature=0,
)
langgraph_llm = ChatOpenAI(
model="YOUR_CHAT_MODEL",
base_url="YOUR_CHAT_BASE_URL",
api_key="YOUR_CHAT_API_KEY",
temperature=0,
)
db_pool = ... #an oracledb connection or connection pool
agent_memory = OracleAgentMemory(
connection=db_pool,
embedder=embedder,
llm=memory_llm,
)
thread_id = "langgraph_short_term_demo"
user_id = "user_123"
agent_id = "assistant_456"
##Configure short term memory middleware and a prebuilt LangGraph agent
def _message_text(message: BaseMessage | Any) -> str:
content = getattr(message, "content", "")
if isinstance(content, str):
return content
return str(content)
def _is_context_card_message(message: BaseMessage) -> bool:
return isinstance(message, HumanMessage) and (
getattr(message, "name", None) == "memory_context_card"
)
class OracleShortTermMemoryMiddleware(AgentMiddleware):
"""Persist LangGraph turns and compact prompts with an OracleAgentMemory context card.
Notes
-----
- ``before_model()`` receives the current LangGraph message state for this turn.
After compaction, that state already includes the synthetic ``memory_context_card``
message returned by a previous ``before_model()`` call.
- The middleware strips that synthetic message back out before persisting or
measuring token usage so OracleAgentMemory only stores real user/assistant turns
and the compaction threshold is based on the organic conversation.
- When compaction triggers, the middleware replaces the message history with one
context-card message plus the most recent raw turns. On the next turn, that
same injected message is seen again and filtered out before recomputing the
next compacted prompt.
"""
def __init__(
self,
memory: OracleAgentMemory,
thread_id: str,
user_id: str,
agent_id: str,
compaction_token_trigger: int,
kept_message_count: int,
) -> None:
self._thread = memory.create_thread(
thread_id=thread_id,
user_id=user_id,
agent_id=agent_id,
context_summary_update_frequency=4,
)
self._compaction_token_trigger = int(compaction_token_trigger)
self._kept_message_count = int(kept_message_count)
self._persisted_message_ids: set[str] = set()
def before_model(
self,
state: dict[str, Any],
runtime: Runtime[Any],
) -> dict[str, Any] | None:
del runtime
messages = list(state["messages"])
#^ This will contain the context card message once the compaction occurs
raw_messages = [message for message in messages if not _is_context_card_message(message)]
self._persist_new_messages(raw_messages)
#we exclude the context card from the token counting
if count_tokens_approximately(raw_messages) < self._compaction_token_trigger:
return None
context_card = self._thread.get_context_card().content
if not context_card:
context_card = "<context_card>\n No relevant short-term context yet.\n</context_card>"
return {
"messages": [
RemoveMessage(id=REMOVE_ALL_MESSAGES), #Clear existing message state.
HumanMessage(content=context_card, name="memory_context_card"),
*raw_messages[-self._kept_message_count :],
]
}
def _persist_new_messages(self, messages: list[BaseMessage]) -> None:
persisted: list[dict[str, str]] = []
for message in messages:
#Persist only the conversational roles that map directly to short-
#term memory turns. Tool/system/synthetic messages are skipped here.
role = (
"user"
if isinstance(message, HumanMessage)
else "assistant" if isinstance(message, AIMessage) else None
)
if role is None:
continue
content = _message_text(message).strip()
if not content:
continue
#LangGraph messages usually have stable IDs. When they do not, fall back
#to a content-derived key so the same turn is not persisted repeatedly if
#the caller reuses the returned message list across later invocations.
message_id = str(getattr(message, "id", "") or f"{role}:{hash(content)}")
if message_id in self._persisted_message_ids:
continue
#Track what this middleware instance has already written so each real turn
#is added to Oracle once even though later turns may still carry the same
#messages in the LangGraph state.
self._persisted_message_ids.add(message_id)
persisted.append({"role": role, "content": content})
if persisted:
self._thread.add_messages(persisted)
short_term_middleware = OracleShortTermMemoryMiddleware(
memory=agent_memory,
thread_id=thread_id,
user_id=user_id,
agent_id=agent_id,
compaction_token_trigger=120,
kept_message_count=3,
)
agent = create_agent(
model=langgraph_llm,
tools=[],
middleware=[short_term_middleware],
)
##Answer later turns with the middleware backed agent
messages: list[BaseMessage] = []
def print_current_context_card(messages: list[BaseMessage]) -> None:
for message in messages:
if _is_context_card_message(message):
print(_message_text(message))
return
print("<context_card>\n No injected context card yet.\n</context_card>")
def run_turn(user_text: str) -> str:
messages.append(HumanMessage(content=user_text))
result = agent.invoke({"messages": messages})
messages[:] = list(result["messages"])
assistant_message = next(
message for message in reversed(messages) if isinstance(message, AIMessage)
)
return _message_text(assistant_message)
run_turn(
"I'm Maya. I'm migrating our nightly invoice reconciliation workflow "
"from cron jobs to LangGraph."
)
run_turn("The failing step right now is ledger enrichment after reconciliation.")
final_answer = run_turn(
"What workflow am I migrating, which step is failing, and who am I?"
)
print_current_context_card(messages)
#<context_card>
#<topics>
#<topic>invoice reconciliation migration</topic>
#<topic>ledger enrichment failure</topic>
#...
#</topics>
#<summary>
#Maya is migrating the nightly invoice reconciliation workflow from cron jobs
#to LangGraph. The failing step is ledger enrichment after reconciliation.
#</summary>
#...
#</context_card>
print(final_answer)
#You're Maya, migrating your nightly invoice reconciliation workflow from cron jobs
#to LangGraph, and the ledger-enrichment step after reconciliation is currently failing.