MemGPT/letta/memory.py

import datetime
from abc import ABC, abstractmethod
from typing import Callable, Dict, List, Tuple, Union

from letta.constants import MESSAGE_SUMMARY_REQUEST_ACK, MESSAGE_SUMMARY_WARNING_FRAC
from letta.embeddings import embedding_model, parse_and_chunk_text, query_embedding
from letta.llm_api.llm_api_tools import create
from letta.prompts.gpt_summarize import SYSTEM as SUMMARY_PROMPT_SYSTEM
from letta.schemas.agent import AgentState
from letta.schemas.memory import Memory
from letta.schemas.message import Message
from letta.schemas.passage import Passage
from letta.utils import (
    count_tokens,
    extract_date_from_timestamp,
    get_local_time,
    printd,
    validate_date_format,
)


def get_memory_functions(cls: Memory) -> Dict[str, Callable]:
    """Get memory functions for a memory class"""
    functions = {}

    # collect base memory functions (should not be included)
    base_functions = []
    for func_name in dir(Memory):
        funct = getattr(Memory, func_name)
        if callable(funct):
            base_functions.append(func_name)

    for func_name in dir(cls):
        if func_name.startswith("_") or func_name in ["load", "to_dict"]:  # skip base functions
            continue
        if func_name in base_functions:  # dont use BaseMemory functions
            continue
        func = getattr(cls, func_name)
        if not callable(func):  # not a function
            continue
        functions[func_name] = func
    return functions


def _format_summary_history(message_history: List[Message]):
    # TODO use existing prompt formatters for this (eg ChatML)
    return "\n".join([f"{m.role}: {m.text}" for m in message_history])


def summarize_messages(
    agent_state: AgentState,
    message_sequence_to_summarize: List[Message],
    insert_acknowledgement_assistant_message: bool = True,
):
    """Summarize a message sequence using GPT"""
    # we need the context_window
    context_window = agent_state.llm_config.context_window

    summary_prompt = SUMMARY_PROMPT_SYSTEM
    summary_input = _format_summary_history(message_sequence_to_summarize)
    summary_input_tkns = count_tokens(summary_input)
    if summary_input_tkns > MESSAGE_SUMMARY_WARNING_FRAC * context_window:
        trunc_ratio = (MESSAGE_SUMMARY_WARNING_FRAC * context_window / summary_input_tkns) * 0.8  # For good measure...
        cutoff = int(len(message_sequence_to_summarize) * trunc_ratio)
        summary_input = str(
            [summarize_messages(agent_state, message_sequence_to_summarize=message_sequence_to_summarize[:cutoff])]
            + message_sequence_to_summarize[cutoff:]
        )

    dummy_user_id = agent_state.user_id
    dummy_agent_id = agent_state.id
    message_sequence = []
    message_sequence.append(Message(user_id=dummy_user_id, agent_id=dummy_agent_id, role="system", text=summary_prompt))
    if insert_acknowledgement_assistant_message:
        message_sequence.append(Message(user_id=dummy_user_id, agent_id=dummy_agent_id, role="assistant", text=MESSAGE_SUMMARY_REQUEST_ACK))
    message_sequence.append(Message(user_id=dummy_user_id, agent_id=dummy_agent_id, role="user", text=summary_input))

    response = create(
        llm_config=agent_state.llm_config,
        user_id=agent_state.user_id,
        messages=message_sequence,
        stream=False,
    )

    printd(f"summarize_messages gpt reply: {response.choices[0]}")
    reply = response.choices[0].message.content
    return reply


class ArchivalMemory(ABC):
    @abstractmethod
    def insert(self, memory_string: str):
        """Insert new archival memory

        :param memory_string: Memory string to insert
        :type memory_string: str
        """

    @abstractmethod
    def search(self, query_string, count=None, start=None) -> Tuple[List[str], int]:
        """Search archival memory

        :param query_string: Query string
        :type query_string: str
        :param count: Number of results to return (None for all)
        :type count: Optional[int]
        :param start: Offset to start returning results from (None if 0)
        :type start: Optional[int]

        :return: Tuple of (list of results, total number of results)
        """

    @abstractmethod
    def compile(self) -> str:
        """Convert archival memory into a string representation for a prompt"""

    @abstractmethod
    def count(self) -> int:
        """Count the number of memories in the archival memory"""


class RecallMemory(ABC):
    @abstractmethod
    def text_search(self, query_string, count=None, start=None):
        """Search messages that match query_string in recall memory"""

    @abstractmethod
    def date_search(self, start_date, end_date, count=None, start=None):
        """Search messages between start_date and end_date in recall memory"""

    @abstractmethod
    def compile(self) -> str:
        """Convert recall memory into a string representation for a prompt"""

    @abstractmethod
    def count(self) -> int:
        """Count the number of memories in the recall memory"""

    @abstractmethod
    def insert(self, message: Message):
        """Insert message into recall memory"""


class DummyRecallMemory(RecallMemory):
    """Dummy in-memory version of a recall memory database (eg run on MongoDB)

    Recall memory here is basically just a full conversation history with the user.
    Queryable via string matching, or date matching.

    Recall Memory: The AI's capability to search through past interactions,
    effectively allowing it to 'remember' prior engagements with a user.
    """

    def __init__(self, message_database=None, restrict_search_to_summaries=False):
        self._message_logs = [] if message_database is None else message_database  # consists of full message dicts

        # If true, the pool of messages that can be queried are the automated summaries only
        # (generated when the conversation window needs to be shortened)
        self.restrict_search_to_summaries = restrict_search_to_summaries

    def __len__(self):
        return len(self._message_logs)

    def count(self) -> int:
        return len(self)

    def compile(self) -> str:
        # don't dump all the conversations, just statistics
        system_count = user_count = assistant_count = function_count = other_count = 0
        for msg in self._message_logs:
            role = msg["message"]["role"]
            if role == "system":
                system_count += 1
            elif role == "user":
                user_count += 1
            elif role == "assistant":
                assistant_count += 1
            elif role == "function":
                function_count += 1
            else:
                other_count += 1
        memory_str = (
            f"Statistics:"
            + f"\n{len(self._message_logs)} total messages"
            + f"\n{system_count} system"
            + f"\n{user_count} user"
            + f"\n{assistant_count} assistant"
            + f"\n{function_count} function"
            + f"\n{other_count} other"
        )
        return f"\n### RECALL MEMORY ###" + f"\n{memory_str}"

    def insert(self, message):
        raise NotImplementedError("This should be handled by the PersistenceManager, recall memory is just a search layer on top")

    def text_search(self, query_string, count=None, start=None):
        # in the dummy version, run an (inefficient) case-insensitive match search
        message_pool = [d for d in self._message_logs if d["message"]["role"] not in ["system", "function"]]
        start = 0 if start is None else int(start)
        count = 0 if count is None else int(count)

        printd(
            f"recall_memory.text_search: searching for {query_string} (c={count}, s={start}) in {len(self._message_logs)} total messages"
        )
        matches = [
            d for d in message_pool if d["message"]["content"] is not None and query_string.lower() in d["message"]["content"].lower()
        ]
        printd(f"recall_memory - matches:\n{matches[start:start+count]}")

        # start/count support paging through results
        if start is not None and count is not None:
            return matches[start : start + count], len(matches)
        elif start is None and count is not None:
            return matches[:count], len(matches)
        elif start is not None and count is None:
            return matches[start:], len(matches)
        else:
            return matches, len(matches)

    def date_search(self, start_date, end_date, count=None, start=None):
        message_pool = [d for d in self._message_logs if d["message"]["role"] not in ["system", "function"]]

        # First, validate the start_date and end_date format
        if not validate_date_format(start_date) or not validate_date_format(end_date):
            raise ValueError("Invalid date format. Expected format: YYYY-MM-DD")

        # Convert dates to datetime objects for comparison
        start_date_dt = datetime.datetime.strptime(start_date, "%Y-%m-%d")
        end_date_dt = datetime.datetime.strptime(end_date, "%Y-%m-%d")

        # Next, match items inside self._message_logs
        matches = [
            d
            for d in message_pool
            if start_date_dt <= datetime.datetime.strptime(extract_date_from_timestamp(d["timestamp"]), "%Y-%m-%d") <= end_date_dt
        ]

        # start/count support paging through results
        start = 0 if start is None else int(start)
        count = 0 if count is None else int(count)
        if start is not None and count is not None:
            return matches[start : start + count], len(matches)
        elif start is None and count is not None:
            return matches[:count], len(matches)
        elif start is not None and count is None:
            return matches[start:], len(matches)
        else:
            return matches, len(matches)


class BaseRecallMemory(RecallMemory):
    """Recall memory based on base functions implemented by storage connectors"""

    def __init__(self, agent_state, restrict_search_to_summaries=False):
        # If true, the pool of messages that can be queried are the automated summaries only
        # (generated when the conversation window needs to be shortened)
        self.restrict_search_to_summaries = restrict_search_to_summaries
        from letta.agent_store.storage import StorageConnector

        self.agent_state = agent_state

        # create embedding model
        self.embed_model = embedding_model(agent_state.embedding_config)
        self.embedding_chunk_size = agent_state.embedding_config.embedding_chunk_size

        # create storage backend
        self.storage = StorageConnector.get_recall_storage_connector(user_id=agent_state.user_id, agent_id=agent_state.id)
        # TODO: have some mechanism for cleanup otherwise will lead to OOM
        self.cache = {}

    def get_all(self, start=0, count=None):
        start = 0 if start is None else int(start)
        count = 0 if count is None else int(count)
        results = self.storage.get_all(start, count)
        results_json = [message.to_openai_dict() for message in results]
        return results_json, len(results)

    def text_search(self, query_string, count=None, start=None):
        start = 0 if start is None else int(start)
        count = 0 if count is None else int(count)
        results = self.storage.query_text(query_string, count, start)
        results_json = [message.to_openai_dict_search_results() for message in results]
        return results_json, len(results)

    def date_search(self, start_date, end_date, count=None, start=None):
        start = 0 if start is None else int(start)
        count = 0 if count is None else int(count)
        results = self.storage.query_date(start_date, end_date, count, start)
        results_json = [message.to_openai_dict_search_results() for message in results]
        return results_json, len(results)

    def compile(self) -> str:
        total = self.storage.size()
        system_count = self.storage.size(filters={"role": "system"})
        user_count = self.storage.size(filters={"role": "user"})
        assistant_count = self.storage.size(filters={"role": "assistant"})
        function_count = self.storage.size(filters={"role": "function"})
        other_count = total - (system_count + user_count + assistant_count + function_count)

        memory_str = (
            f"Statistics:"
            + f"\n{total} total messages"
            + f"\n{system_count} system"
            + f"\n{user_count} user"
            + f"\n{assistant_count} assistant"
            + f"\n{function_count} function"
            + f"\n{other_count} other"
        )
        return f"\n### RECALL MEMORY ###" + f"\n{memory_str}"

    def insert(self, message: Message):
        self.storage.insert(message)

    def insert_many(self, messages: List[Message]):
        self.storage.insert_many(messages)

    def save(self):
        self.storage.save()

    def __len__(self):
        return self.storage.size()

    def count(self) -> int:
        return len(self)


class EmbeddingArchivalMemory(ArchivalMemory):
    """Archival memory with embedding based search"""

    def __init__(self, agent_state: AgentState, top_k: int = 100):
        """Init function for archival memory

        :param archival_memory_database: name of dataset to pre-fill archival with
        :type archival_memory_database: str
        """
        from letta.agent_store.storage import StorageConnector

        self.top_k = top_k
        self.agent_state = agent_state

        # create embedding model
        self.embed_model = embedding_model(agent_state.embedding_config)
        if agent_state.embedding_config.embedding_chunk_size is None:
            raise ValueError(f"Must set {agent_state.embedding_config.embedding_chunk_size}")
        else:
            self.embedding_chunk_size = agent_state.embedding_config.embedding_chunk_size

        # create storage backend
        self.storage = StorageConnector.get_archival_storage_connector(user_id=agent_state.user_id, agent_id=agent_state.id)
        # TODO: have some mechanism for cleanup otherwise will lead to OOM
        self.cache = {}

    def create_passage(self, text, embedding):
        return Passage(
            user_id=self.agent_state.user_id,
            agent_id=self.agent_state.id,
            text=text,
            embedding=embedding,
            embedding_config=self.agent_state.embedding_config,
        )

    def save(self):
        """Save the index to disk"""
        self.storage.save()

    def insert(self, memory_string, return_ids=False) -> Union[bool, List[str]]:
        """Embed and save memory string"""

        if not isinstance(memory_string, str):
            raise TypeError("memory must be a string")

        try:
            passages = []

            # breakup string into passages
            for text in parse_and_chunk_text(memory_string, self.embedding_chunk_size):
                embedding = self.embed_model.get_text_embedding(text)
                # fixing weird bug where type returned isn't a list, but instead is an object
                # eg: embedding={'object': 'list', 'data': [{'object': 'embedding', 'embedding': [-0.0071973633, -0.07893023,
                if isinstance(embedding, dict):
                    try:
                        embedding = embedding["data"][0]["embedding"]
                    except (KeyError, IndexError):
                        # TODO as a fallback, see if we can find any lists in the payload
                        raise TypeError(
                            f"Got back an unexpected payload from text embedding function, type={type(embedding)}, value={embedding}"
                        )
                passages.append(self.create_passage(text, embedding))

            # grab the return IDs before the list gets modified
            ids = [str(p.id) for p in passages]

            # insert passages
            self.storage.insert_many(passages)

            if return_ids:
                return ids
            else:
                return True

        except Exception as e:
            print("Archival insert error", e)
            raise e

    def search(self, query_string, count=None, start=None):
        """Search query string"""
        start = 0 if start is None else int(start)
        count = self.top_k if count is None else int(count)

        if not isinstance(query_string, str):
            return TypeError("query must be a string")

        try:
            if query_string not in self.cache:
                # self.cache[query_string] = self.retriever.retrieve(query_string)
                query_vec = query_embedding(self.embed_model, query_string)
                self.cache[query_string] = self.storage.query(query_string, query_vec, top_k=self.top_k)

            end = min(count + start, len(self.cache[query_string]))

            results = self.cache[query_string][start:end]
            results = [{"timestamp": get_local_time(), "content": node.text} for node in results]
            return results, len(results)
        except Exception as e:
            print("Archival search error", e)
            raise e

    def compile(self) -> str:
        limit = 10
        passages = []
        for passage in list(self.storage.get_all(limit=limit)):  # TODO: only get first 10
            passages.append(str(passage.text))
        memory_str = "\n".join(passages)
        return f"\n### ARCHIVAL MEMORY ###" + f"\n{memory_str}" + f"\nSize: {self.storage.size()}"

    def __len__(self):
        return self.storage.size()

    def count(self) -> int:
        return len(self)