Proposal: Adding Slack Toolkit

[j-tara]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

We would like to integrate the Slack toolkit into Langchain JS, which would have similar functionality to the Python Langchain repository Slack toolkit. This integration will support developers in building custom AI applications with agents that can perform operations on Slack using the following methods from the Node Slack SDK WebAPI:

• conversations.info: Retrieve metadata about specific conversations, which agents can use to understand context or member information
• search.messages: Enable agents to search messages across Slack channels, providing access to valuable insights and context within Slack
• chat.scheduleMessage: Allow agents to schedule messages, helping them manage timed interactions within Slack channels
• chat.postMessage: Let agents post messages in real time, enhancing their ability to interact seamlessly in Slack

Motivation

Adding a Slack toolkit to Langchain JS is valuable because it aligns the JavaScript implementation with the Python Langchain repository, offering a consistent developer experience across both languages. It will allow developers using Langchain JS to easily integrate the Slack SDK into their AI applications, enabling agents to automate conversational tasks, schedule messages, and retrieve information. With this toolkit, Slack-driven AI solutions can be more accessible in the JavaScript ecosystem.

Proposal (If applicable)

We are a team of five students at the University of Toronto, aiming to submit a pull request no later than mid November.

Implementation Outline

We plan to add the following files for integrating the Slack toolkit:

• libs/langchain-community/src/tools/slack.ts
• libs/langchain-community/src/tools/tests/slack.int.test.ts

Here is a sample implementation of SlackGetMessageTool() in slack.ts:

import {
    conversations.info,
    search.messages,
    chat.scheduleMessage,
    chat.postMessage,
  } from "@slack/web-api";
  import { getEnvironmentVariable } from "@langchain/core/utils/env";
  import { Tool } from "@langchain/core/tools";
 
  /**
   * Base tool parameters for the Slack tools
   */
  interface SlackToolParams {
    token?: string;
    client?: Client;
  }
 
 
  /**
   * A tool for retrieving messages from a slack channel using a bot.
   * It extends the base Tool class and implements the _call method to
   * perform the retrieve operation. Requires a slack token which can be set
   * in the environment variables.
   * The _call method takes the search query as the input argument.
   * It returns the messages including the channel and text.
   */
  export class SlackGetMessagesTool extends Tool {
    static lc_name() {
      return "SlackGetMessagesTool";
    }
 
    name = "slack-get-messages";
 
    description = `A slack tool. useful for reading messages from slack channels.
    Input should be the search query.`;
 
    protected token: string;
 
    protected client: Client;
 
    constructor(fields?: SlackToolParams) {
      super();
 
      const {
        token = getEnvironmentVariable("SLACK_TOKEN"),
        client,
      } = fields ?? {};
 
      if (!token) {
        throw new Error(
          "Environment variable SLACK_TOKEN missing, but is required for SlackGetMessagesTool."
        );
      }
     
      this.client =
        client ??
        new WebClient(token);


      this.token = token;
    }
 
    /** @ignore */
    async _call(searchTerm: string): Promise<string> {
      try {


        const results = await this.client.getMessage({
            query: searchTerm
        })
 
        return JSON.stringify(results);
      } catch (err) {
        return "Error getting messages.";
      }
    }
  }

Example Usage

npm install @langchain/openai @langchain/core
import {
 SlackGetChannelTool,
 SlackGetMessageTool,
 SlackScheduleMessageTool,
 SlackSendMessageTool
} from "@langchain/community/tools/slack";

// Get channel details given channel ID
const getChannelTool = new SlackGetChannelTool();
const channelResults = await getChannelTool.invoke("0123456789");
console.log(channelResults);

// Search and get message
const getMessageTool = new SlackGetMessageTool();
const messageResults = await getMessageTool.invoke("Hi");
console.log(messageResults);

// Schedule message
const scheduleTool = new SlackScheduleMessageTool();
const scheduleResults = await scheduleTool.invoke({channel: “0123456789”, message: “Hi”, post_at: “299876400”});
console.log(scheduleResults);

// Send message
const sendMessageTool = new SlackSendMessageTool();
const sendMessageResults = await sendMessageTool.invoke({channel: “0123456789”, text: “Test Message”});
console.log(sendMessageResults);

[j-tara]

Hey @jacoblee93 @bracesproul 👋we are a group of students at the University of Toronto willing to work on this issue, helping to bring more python Langchain features to Langchain JS. We were wondering if this is something that we can create an issue for. Thank you!

[j-tara]

Hi @jacoblee93! We started working on the code for this feature, and we were wondering if this could be made into an issue. We'd love to hear your thoughts on it. Thank you!

[j-tara]

@jacoblee93 We put up a PR for the Slack tool here. Whenever you have a chance to look at it, it is greatly appreciated!