AWSTemplateFormatVersion: '2010-09-09'
Transform: AWS::Serverless-2016-10-31
Description: >
    slack-gpt-backend    

# More info about Globals: https://github.com/awslabs/serverless-application-model/blob/master/docs/globals.rst
Globals:
    Function:
        Timeout: 120
        MemorySize: 128
        Environment:
            Variables:
                LOG_LEVEL: INFO
                POWERTOOLS_LOGGER_SAMPLE_RATE: 0.1
                POWERTOOLS_LOGGER_LOG_EVENT: true
                POWERTOOLS_SERVICE_NAME: slack-gpt-backend

Parameters:
    OpenAiApiKey:
        Type: AWS::SSM::Parameter::Value<String>
        Default: '/slack-gpt-backend/OpenAiApiKey'
    SlackToken:
        Type: AWS::SSM::Parameter::Value<String>
        Default: '/slack-gpt-backend/SlackToken'
    SlackChannel:
        Type: AWS::SSM::Parameter::Value<String>
        Default: '/slack-gpt-backend/SlackChannel'
    SlackContextTeamId:
        Type: AWS::SSM::Parameter::Value<String>
        Default: '/slack-gpt-backend/SlackContextTeamId'
    SlackUserId:
        Type: AWS::SSM::Parameter::Value<String>
        Default: '/slack-gpt-backend/SlackUserId'
    SlackReplyUsername:
        Type: AWS::SSM::Parameter::Value<String>
        Default: '/slack-gpt-backend/SlackReplyUsername'

Resources:
    postMessageFunction:
        Type: AWS::Serverless::Function
        Properties:
            CodeUri: src/
            Handler: app.lambda_handler
            Runtime: python3.9
            Layers:
                - !Sub arn:aws:lambda:${AWS::Region}:017000801446:layer:AWSLambdaPowertoolsPython:21
            Policies:
                - DynamoDBCrudPolicy:
                      TableName: !Ref ConversationHistoriesTable
            Environment:
                Variables:
                    CONVERSATIONS_HISTORIES_TABLE: !Ref ConversationHistoriesTable
                    OPENAI_API_KEY: !Ref OpenAiApiKey
                    SLACK_TOKEN: !Ref SlackToken
                    SLACK_USER_ID: !Ref SlackUserId
                    SLACK_CHANNEL: !Ref SlackChannel
                    SLACK_CONTEXT_TEAM_ID: !Ref SlackContextTeamId
                    SLACK_REPLY_USERNAME: !Ref SlackReplyUsername
            Architectures:
                - x86_64
            Description: Slackで入力されたメッセージを受け取り、OpenAI APIを通して得た返答をSlackのメッセージとして投稿します。
            Events:
                Api:
                    Type: Api
                    Properties:
                        Path: /{proxy}
                        Method: ANY
    ConversationHistoriesTable:
        Type: AWS::Serverless::SimpleTable
        Properties:
            PrimaryKey:
                Name: id
                Type: String
    ApplicationResourceGroup:
        Type: AWS::ResourceGroups::Group
        Properties:
            Name:
                Fn::Join:
                    - ''
                    - - ApplicationInsights-SAM-
                      - Ref: AWS::StackName
            ResourceQuery:
                Type: CLOUDFORMATION_STACK_1_0
    ApplicationInsightsMonitoring:
        Type: AWS::ApplicationInsights::Application
        Properties:
            ResourceGroupName:
                Fn::Join:
                    - ''
                    - - ApplicationInsights-SAM-
                      - Ref: AWS::StackName
            AutoConfigurationEnabled: 'true'
        DependsOn: ApplicationResourceGroup

Outputs:
    WebEndpoint:
        Description: API Gateway endpoint URL for Prod stage
        Value: !Sub 'https://${ServerlessRestApi}.execute-api.${AWS::Region}.amazonaws.com/Prod/'

from langchain.prompts import PromptTemplate
from langchain import OpenAI, ConversationChain
from langchain.chains.conversation.memory import ConversationSummaryBufferMemory
from slack_sdk import WebClient
from slack_sdk.errors import SlackApiError
import os
import boto3

reply_llm = OpenAI(temperature=0, max_tokens=500, client=None)
summary_llm = OpenAI(temperature=0, max_tokens=1000, client=None)
conversation_template = """以下は、私とAIが仲良く会話している様子です。AIは饒舌で、その文脈から具体的な内容をたくさん教えてくれます。AIは質問に対する答えを知らない場合、正直に「知らない」と答えます。

{history}
私: {input}
AI:"""

conversation_prompt = PromptTemplate(
    input_variables=["history", "input"], template=conversation_template
)

summary_template = """会話内容を順次要約し、前回の要約に追加して新たな要約を返してください。

### 現在の要約

{summary}

### 新しい会話

{new_lines}

### 新しい要約

"""

slack_token = os.environ["SLACK_TOKEN"]
slack_client = WebClient(token=slack_token)
dynamodb = boto3.resource("dynamodb")
table = dynamodb.Table(os.environ["CONVERSATIONS_HISTORIES_TABLE"])
summary_prompt = PromptTemplate(
    input_variables=["summary", "new_lines"], template=summary_template
)
memory = ConversationSummaryBufferMemory(
    human_prefix="私", llm=summary_llm, max_token_limit=2000, prompt=summary_prompt
)
conversation = ConversationChain(
    llm=reply_llm, prompt=conversation_prompt, memory=memory, verbose=False
)


def get_reply(message, buffer=[], summary_buffer=""):
    conversation.memory.buffer = buffer
    conversation.memory.moving_summary_buffer = summary_buffer
    reply = conversation.predict(input=message)

    return reply, conversation.memory


def post_messsage2slack(channel, message, reply_username):
    result = slack_client.chat_postMessage(
        channel=channel, text=message, username=reply_username
    )

    return result


def save_context(context_team_id, channel, buffer, summary_buffer):
    item = {
        "id": f"{context_team_id}/{channel}",
        "buffer": buffer,
        "summary_buffer": summary_buffer,
    }
    table.put_item(Item=item)


def load_context(context_team_id, channel):
    response = table.get_item(
        Key={
            "id": f"{context_team_id}/{channel}",
        }
    )

    return response

from aws_lambda_powertools import Logger
from aws_lambda_powertools.utilities.typing import LambdaContext
from aws_lambda_powertools.event_handler import APIGatewayRestResolver
from aws_lambda_powertools.logging import correlation_paths
from backend import get_reply, save_context, load_context, post_messsage2slack
import os

logger = Logger()
app = APIGatewayRestResolver()
slack_user_id = os.environ["SLACK_USER_ID"]
slack_channel = os.environ["SLACK_CHANNEL"]
slack_context_team_id = os.environ["SLACK_CONTEXT_TEAM_ID"]
slack_reply_username = os.environ["SLACK_REPLY_USERNAME"]


@app.post("/events")
def post_message():
    event = app.current_event

    # Slackからは複数回呼び出されるため、最初の呼び出し以外は無視するようにします。
    retry_counts = event.get("multiValueHeaders", {}).get("X-Slack-Retry-Num", [0])

    if retry_counts[0] != 0:
        logger.info(f"Skip slack retrying({retry_counts}).")
        return {}

    body = app.current_event.json_body

    # Slack Appのイベントハンドラーとして登録する際の対応です。
    if "challenge" in body:
        return {"challenge": body["challenge"]}

    logger.info(body)

    message = body["event"]["text"]
    # 対象にメンションされたイベント以外に反応すると無限ループになるので要注意です。
    if not message.startswith(slack_user_id):
        logger.info("Not mentioned me.")
        return {}

    history = load_context(slack_context_team_id, slack_channel).get("Item", {})

    logger.info(history)

    buffer = history.get("buffer", [])
    summary_buffer = history.get("summary_buffer", "")

    reply, memory = get_reply(message, buffer, summary_buffer)

    logger.info(reply)

    save_context(
        slack_context_team_id,
        slack_channel,
        memory.buffer,
        memory.moving_summary_buffer,
    )

    post_messsage2slack(slack_channel, reply, slack_reply_username)

    return {}

@logger.inject_lambda_context(correlation_id_path=correlation_paths.API_GATEWAY_REST)
def lambda_handler(event: dict, context: LambdaContext) -> dict:
    return app.resolve(event, context)

sam build
sam deploy --guided

sam deploy