How to Summarize Slack Threads with OpenAI using Pipedream

Dashboard > New > Workflow

Create New Pipedream Workflow

Go to pipedream.com and click New in your dashboard. You'll see a blank workflow builder. This workflow needs to start with a Slack trigger that detects emoji reactions. The trigger will fire instantly when someone adds the 📝 emoji to any message.

1. 1Click the purple 'New' button in top left
2. 2Select 'Workflow' from the dropdown
3. 3Choose 'Start with a trigger'
4. 4Search for 'Slack' in the app list

Trigger > Slack > Reaction Added

Configure Slack Reaction Added Trigger

Select the 'Reaction Added' event from Slack's trigger options. This fires when any emoji gets added to messages in channels where your bot has access. You'll need to connect your Slack workspace and configure which channels to monitor. The trigger captures the reaction details, message content, and thread context.

1. 1Select 'Reaction Added' from the event list
2. 2Click 'Connect Account' and authorize your Slack workspace
3. 3Leave 'Channel' blank to monitor all channels
4. 4Click 'Save' to finalize the trigger

Steps > Add Step > Code

Add Filter for 📝 Emoji

Add a code step that checks if the reaction is specifically the 📝 emoji. Slack sends all emoji reactions through this trigger, so you need to filter for only the notebook emoji. The code examines the reaction name and stops execution if it's not the target emoji.

1. 1Click the + button below your trigger
2. 2Select 'Run custom code'
3. 3Choose 'Node.js' as the runtime
4. 4Paste the emoji filter code in the editor

Steps > Add Step > Slack > Get Thread Messages

Fetch Complete Thread Messages

Add another Slack step to retrieve all messages in the thread. The reaction trigger only gives you the original message, but you need the full conversation for summarization. Use Slack's conversations.replies API to get the complete thread including timestamps and user information.

1. 1Click + to add another step
2. 2Search for Slack and select it
3. 3Choose 'Get Thread Messages' action
4. 4Map the channel ID and thread timestamp from the trigger

Steps > Add Step > Code

Format Thread Content for OpenAI

Create a code step that converts the raw Slack thread data into clean text for the AI model. Remove Slack formatting, resolve user mentions to readable names, and structure the conversation chronologically. This step is critical for getting quality summaries from GPT.

1. 1Add a new Node.js code step
2. 2Access the thread messages from the previous step
3. 3Loop through messages to build formatted text
4. 4Export the formatted content for the next step

Steps > Add Step > OpenAI > Chat Completions

Connect OpenAI Account

Add an OpenAI step and connect your API account. You'll need an OpenAI API key from platform.openai.com. Choose the Chat Completions action to send the formatted thread to GPT-4 or GPT-3.5-turbo for summarization.

1. 1Click + to add a step and select OpenAI
2. 2Choose 'Chat Completions' from the actions
3. 3Click 'Connect Account' and paste your API key
4. 4Select 'gpt-4' or 'gpt-3.5-turbo' as the model

OpenAI Step > Messages Configuration

Configure Summary Prompt

Set up the system and user prompts to get consistent, useful summaries. The system prompt defines the AI's role and output format. The user prompt contains the actual thread content. Specify that you want 2-3 sentences focusing on decisions made and action items.

1. 1Set Role to 'system' for the first message
2. 2Enter the system prompt defining summary format
3. 3Add a second message with Role 'user'
4. 4Map the formatted thread content from step 5

Steps > Add Step > Slack > Send Message

Post Summary Back to Slack

Add a final Slack step to post the AI summary back to the original channel as a thread reply. Use the 'Send Message' action and make sure to set the thread timestamp so it appears as a reply, not a new message. Include some visual formatting to make the summary stand out.

1. 1Add a Slack 'Send Message to Channel' step
2. 2Map the channel ID from the original trigger
3. 3Set thread_ts to the original message timestamp
4. 4Map the OpenAI summary to the message text

Workflow > Deploy > Test in Slack

Test the Complete Workflow

Deploy your workflow and test it end-to-end in Slack. Go to any channel thread and react with 📝. The workflow should trigger, fetch the thread, generate a summary, and post it back within 10-15 seconds. Check the execution logs in Pipedream to debug any issues.

1. 1Click 'Deploy' in the top right of your workflow
2. 2Go to any Slack thread in your workspace
3. 3React to a message with the 📝 emoji
4. 4Wait 10-15 seconds for the summary to appear

This code formats the raw Slack thread data into clean text for OpenAI and handles user mention resolution. Paste this into your thread formatting code step.

export default defineComponent({
  async run({ steps, $ }) {
    const messages = steps.get_thread_messages.messages || [];

export default defineComponent({
  async run({ steps, $ }) {
    const messages = steps.get_thread_messages.messages || [];
    
    let formattedThread = "";
    
    for (const msg of messages) {
      // Skip bot messages and system messages
      if (msg.bot_id || msg.subtype) continue;
      
      // Get user display name
      const userInfo = await $.send.http({
        method: 'GET',
        url: `https://slack.com/api/users.info`,
        headers: {
          'Authorization': `Bearer ${auths.slack.oauth_access_token}`
        },
        params: { user: msg.user }
      });
      
      const displayName = userInfo.user?.real_name || userInfo.user?.name || 'Unknown User';
      
      // Clean up message text - remove user mentions and links
      let cleanText = msg.text
        .replace(/<@U[A-Z0-9]+>/g, '@user')
        .replace(/<#C[A-Z0-9]+\|([^>]+)>/g, '#$1')
        .replace(/<([^>]+)>/g, '$1');
      
      // Format timestamp
      const timestamp = new Date(parseFloat(msg.ts) * 1000).toLocaleTimeString();
      
      formattedThread += `[${timestamp}] ${displayName}: ${cleanText}\n`;
    }
    
    return { formatted_thread: formattedThread.trim() };
  }
});