Controlling Read and Write to Memory

You can use the memory_access_policy metadata parameter to control a conversation's read from and write to memory.

Allowed values for the memory_access_policy metadata parameter are:

recall_and_store: enable both store and recall (default)
recall_only: this conversation can recall memory but can't store new memory
store_only: this conversation can store memory but can't recall memory
none: neither store nor recall

Example code:

# first conversation
conversation1 = client.conversations.create(
    metadata={ 
		"memory_subject_id": "user_123456",
        "memory_access_policy": "store_only",
	},
)

# a turn on first conversation, this conversation can store memory, but cannot recall memory
response = client.responses.create(
    model="openai.gpt-4.1",
    input="I like Fish. I don't like Shrimp.", 
    conversation=conversation1.id
)
print(response.output_text)

# delay for long-term memory processing
time.sleep(10)

# second conversation, this conversation can recall memory, but cannot store new memory
conversation2 = client.conversations.create(
    metadata={
		"memory_subject_id": "user_123456"
    	"memory_access_policy": "recall_only",
	},
)

# a turn on second conversation
response = client.responses.create(
    model="openai.gpt-4.1",
    input="What do I like",
    conversation=conversation2.id
)

print(response.output_text)

Oracle Cloud Infrastructure Documentation

Controlling Read and Write to Memory