Add Memory

Add new memory items from a list of messages.

Parameters

• user_id (str, required) — User identifier for the memory.

• agent_id (str, required) — Agent identifier for the memory.

• messages (list[Message], required) — List of messages to store in memory. Each Message follows the GL SDK format and contains a role and content.

• scopes (set[MemoryScope], optional) — Specifies the target for memory management, either for a specific participant or for combined access (USER or ASSISTANT). Default is [USER].

  • USER: Writes and retrieves memories targeting messages from user.

  • ASSISTANT: Writes and retrieves memories targeting messages from assistant.

  • USER, ASSISTANT: Writes and retrieves memories targeting messages from both the user and assistant.

• metadata (dict[str, str], optional) — Extra metadata linked to the memory, which can be used to store any supplementary details or context related to it. Default is None.

• infer (boolean, optional) — Whether to infer the memories or directly store the messages. Default is True.

Returns

list[Chunk] — A list of memory chunks created. Each Chunk follows the GL SDK format and includes an id, content, metadata, and score.

Example

import asyncio
from gllm_inference.schema.message import Message
from gllm_memory import MemoryManager
from gllm_memory.enums import MemoryScope

# Initialize
memory_manager = MemoryManager()

messages = [
    Message.user("I love pizza and Italian food"),
    Message.assistant("I'll remember that you love pizza and Italian food"),
    Message.user("My favorite programming language is Python")
]

user_id = "user_123"
agent_id = "agent_xyz"

result = asyncio.run(memory_manager.add(
    user_id=user_id,
    agent_id=agent_id,
    scopes=[MemoryScope.USER],
    messages=messages,
))

print(result)

Expected Output

[
    Chunk(
        id=003634aa-c99c-4617-8f8d-e76aca07511b,
        content=User Prefers pizza and Italian food,
        metadata={
            'user_id': '1f2ef5ee8771...',
            'app_id': 'default-project',
            'metadata': {
                'source': '1f2ef5ee8771...',
                'target': 'agent_general-purpose'
            },
            'categories': None,
            'created_at': '2025-11-04T07:51:26-08:00',
            'updated_at': '2025-11-04T07:51:26-08:00',
            'expiration_date': None,
            'structured_attributes': {
                'year': 2025, 'month': 11, 'day': 4, 'hour': 15, 'minute': 51,
                'day_of_week': 'tuesday', 'week_of_year': 45, 'day_of_year': 308,
                'quarter': 4, 'is_weekend': False
            },
            'event': 'ADD',
        },
        score=None
    ),
    Chunk(
        id=e66f0832-2447-4c5b-ba5f-8fa85957f436,
        content=User's favorite programming language is Python,
        metadata={
            'agent_id': '1f2ef5ee8771...',
            'app_id': 'default-project',
            'metadata': {
                'source': 'agent_general-purpose',
                'target': '1f2ef5ee8771...'
            },
            'categories': None,
            'created_at': '2025-11-04T07:51:37-08:00',
            'updated_at': '2025-11-04T07:51:37-08:00',
            'expiration_date': None,
            'structured_attributes': {
                'year': 2025, 'month': 11, 'day': 4, 'hour': 15, 'minute': 51,
                'day_of_week': 'tuesday', 'week_of_year': 45, 'day_of_year': 308,
                'quarter': 4, 'is_weekend': False
            },
            'event': 'ADD',
        },
        score=None
    )
]