langchain/memory/token_buffer.py

from typing import Any, Dict, List

from langchain_core.language_models import BaseLanguageModel
from langchain_core.messages import BaseMessage, get_buffer_string

from langchain.memory.chat_memory import BaseChatMemory


class ConversationTokenBufferMemory(BaseChatMemory):
    """Conversation chat memory with token limit."""

    human_prefix: str = "Human"
    ai_prefix: str = "AI"
    llm: BaseLanguageModel
    memory_key: str = "history"
    max_token_limit: int = 2000

    @property
    def buffer(self) -> Any:
        """String buffer of memory."""
        return self.buffer_as_messages if self.return_messages else self.buffer_as_str

    @property
    def buffer_as_str(self) -> str:
        """Exposes the buffer as a string in case return_messages is False."""
        return get_buffer_string(
            self.chat_memory.messages,
            human_prefix=self.human_prefix,
            ai_prefix=self.ai_prefix,
        )

    @property
    def buffer_as_messages(self) -> List[BaseMessage]:
        """Exposes the buffer as a list of messages in case return_messages is True."""
        return self.chat_memory.messages

    @property
    def memory_variables(self) -> List[str]:
        """Will always return list of memory variables.

        :meta private:
        """
        return [self.memory_key]

    def load_memory_variables(self, inputs: Dict[str, Any]) -> Dict[str, Any]:
        """Return history buffer."""
        return {self.memory_key: self.buffer}

    def save_context(self, inputs: Dict[str, Any], outputs: Dict[str, str]) -> None:
        """Save context from this conversation to buffer. Pruned."""
        super().save_context(inputs, outputs)
        # Prune buffer if it exceeds max token limit
        buffer = self.chat_memory.messages
        curr_buffer_length = self.llm.get_num_tokens_from_messages(buffer)
        if curr_buffer_length > self.max_token_limit:
            pruned_memory = []
            while curr_buffer_length > self.max_token_limit:
                pruned_memory.append(buffer.pop(0))
                curr_buffer_length = self.llm.get_num_tokens_from_messages(buffer)