Slack App Plus LLM

• Build a Slack App to capture and handle Slack events. Video Part 1
• Integration with OpenAI & Azure OpenAI with support of function call, Assistant API, and Chat Completion API. Video Part 2

How to Setup and Use

1. Create a Slack App

• Go to your Slack workspace to create an App from scratch
  • Refer: Step1 in Slack Quick Start
• Go to Your Apps >> Setting >> Basic Information >> App Credential, note down App ID, Verification Token
• Go to OAuth & Permissions
  • Under Scopes >> Bot Token Scopes, add/remove based on what lambda event handler logic performs, E.g.
    • app_mention:read, channels:history, groups:history, im:history, chat:write
    • Refer: Slack Scope List for Bot Token
  • Click Install to Workspace and note down Bot User OAuth Token
• Go to App Home >> Show Tabs >> messages tab, toggle on and tick Allow users to send Slash commands and messages from the messages tab.

2. Setup infra on AWS using TF

• Configure various TF vars as needed
  • Refer to your.auto.tfvars.sample
  • Refer: How to set TF vars using environment vars
• Run terraform init, plan and apply
• Note down TF output msg_receiver_lambda_function_url

3. Configure Slack App

• Go to the created Slack App >> Event Subscriptions >> Enable Events, toggle on & input msg_receiver_lambda_function_url into Request URL
• Go to Subscribe to Bot Event >> Add Bot User Event, add event types you would like to receive and handle. E.g.
  • app_mention, message.group, message.channel, message.im
  • Refer: Slack Event List
• Go to OAuth & Permissions, click Reinstall to Workspace

4. Interact with Slack App

• EITHER Add this Slack App to Channels you would like to monitor
• OR message this Slack App directly
• (Usage also depends on this Slack App's Event Subscription and Permission setup)

Customize Handling of Slack Events

• Refer to sample_handler() as an example
  • Create additional handler logic e.g. xxx_handler.py
  • Update lambda_function.py

Customize LLM Function Call

• You may implement more function call in openai_handler.py or az_openai_handler.py
• Refer to llm_tools.py as an example

Key Components

• Slack App (create on Slack)

  • To emit Slack event
  • (optional) To perform additional work as part of customizable event handling logic

• Lambda Event Recevier (Setup on AWS)

  • As destination of Slack event subscription
  • Using lambda function url as endpoint
  • Push Slack events to an SQS for further handling
  • Verify initial Slack event subscription challenge
  • Includes necessary IAM role, policy and etc

• SQS (Setup on AWS)

  • SQS to store received Slack events
  • Trigger Lambda handler for processing
  • Includes necessary IAM role, policy, kms and etc

• Lambda Event Handler (Setup on AWS)

  • Process logic to handle Slack events from SQS
    • Ex1. sample_handler: echo the message received
    • Ex2. tag_user_handler: Extract User emails in slack message, find the Slack user, and send a reply message to @user
  • Handling logic should be customized based on needs

• DynamoDB for mapping of Slack thread and OpenAI Assistant thread (Setup on AWS)

  • Store the OpenAI assistant thread id for each Slack thread
  • Only needed for OpenAI Assistant API handler

• DynamoDB to keep thread history (Setup on AWS)

  • As chat completion API requires full message history for each API call, the DDB keeps the messages in each thread
  • Only for OpenAI Chat Completion API handler

Illustration

Further Enhancements

• Replace deprecating Slack verification token with signning secret
• Further protection for lambda function url and alternative setup to expose lambda
• Add handler for LLM self hosted on Ollama