agent-setup-lib v1.0.12

import { TavilySearchResults } from "@langchain/community/tools/tavily_search";
import { CheerioWebBaseLoader } from "@langchain/community/document_loaders/web/cheerio";
import { tool } from "@langchain/core/tools";
import { z } from "zod";

export const tavilyTool = new TavilySearchResults({apiKey: "tavily api key"});

export const webLoader = tool(async (input) => {
  const loader = new CheerioWebBaseLoader(input.url);
  const docs = await loader.load();
  const formattedDocs = docs.map(
    (doc) =>
      `<Document name="${doc.metadata?.title}">\n${doc.pageContent}\n</Document>`,
  );
  return formattedDocs.join("\n\n");
  },
  {
    name: "webpage_loader",
    description: "Scrape the contents of a webpage.",
    schema: z.object({
      url: z.string(),
    }),
  }
)

//////////////////////////////////////////////////////////////////////////////////////////////

// demo.ts
import Llm from "../src/utils/llm";
import { MongoHandler } from "../src/utils/memory";
import { AgentSetup, AgentState } from "../src/agent.setup";
import { ChainCreator, Member } from "../src/index";
import { HumanMessage } from "@langchain/core/messages";
import { createReactAgent } from "@langchain/langgraph/prebuilt";
import { webLoader, tavilyTool } from "./tools/tools";

// Replace these configuration values with your actual keys and endpoints.
const llmConfig = {
  OLLAMA_BASE_URL: "http://localhost:11434",
  OPENAI_CHAT_API_KEY: "your-openai-api-key",
  OPENAI_EMBEDDINGS_API_KEY: "your-openai-api-key",
  ANTHROPIC_API_KEY: "your-anthropic-api-key",
};

async function demo() {
  try {
    

    // 1. Initialize the LLM instance.
    const llm = new Llm(llmConfig);
    const chatLlm = llm.getChatLlm({
      provider: "openai",
      model: "gpt-4",
      temperature: 0.7,
    });
    const embeddings = llm.getEmbeddings({
      provider: "openai",
      model: "text-embedding-ada-002",
    });
    console.log("LLM and embeddings instances created.");




    // 2. Setup the MongoDB memory handler.
    const mongoHandler = new MongoHandler(
      embeddings.instance,
      "mongodb://localhost:27017",
      {
        dbName: "multiAgentDB",
        collectionName: "vectorStore",
        indexName: "defaultIndex",
        textKey: "pageContent",
        embeddingKey: "embedding",
      }
    );
    console.log("MongoDB memory handler initialized.");




    // 3. Create an initial agent state.
    const initialState = {
      messages: [
        new HumanMessage({
          content: "Hello, starting multi-agent workflow.",
        }),
      ],
      team_members: [],
      next: "supervisor",
      instructions: "Select the correct team member for the task",
      stateOption: "",
    };




    // 4. Initialize AgentSetup with the agent state and the chat LLM instance.
    const agentSetup = new AgentSetup(initialState, chatLlm.instance);
    console.log("Agent setup created.");

    async function webSearchNode (state: typeof AgentState.State) {
      const stateModifier = agentSetup.agentStateModifier(
          "You are a web search agent",
          [tavilyTool],
          state.team_members ?? ["Web_Search"],
      )
      const agent = createReactAgent({
          llm: chatLlm.instance,
          tools: [tavilyTool],
          messageModifier: stateModifier(state),
      })
      return agentSetup.runAgentNode({ state, agent: agent, name: "Web_Search" });
    }




    // 5. Define team members with their associated functions and classification.
    const teamMembers: Member[] = [
      {
        name: "Web_Search",
        func: webSearchNode,
        classification: ["Search the web", "look for intresting facts about a topic"]
      }
    ];



    // 6. Create a team supervisor agent using AgentSetup.
    // The supervisor uses the list of team member names to decide the next step.
    //const members:Member[] = teamMembers.map((member) => ({ name: member.name , classification: member.classification}))
    const supervisorAgent = await agentSetup.createTeamSupervisor(
        teamMembers
    );
    console.log("Team supervisor created.");

    // 7. Initialize a checkpointer (saver) using MongoDB.
    const saver = await mongoHandler.initSaver("multiAgentDB", "checkpoints");
    console.log("Checkpointer initialized.");

    // 8. Create the agent chain using ChainCreator.
    const chainCreator = new ChainCreator(saver, supervisorAgent, teamMembers);
    const chain = chainCreator.createChain();
    console.log("Chain created successfully:");

    const finalState = await chain.stream({
        messages: [new HumanMessage("return my last request")],
      }, {
        recursionLimit: 100,
        configurable: { thread_id: "1" }
      });
      for await (const output of finalState) {
        if (!output?.__end__) {
            console.log(output);
            console.log("----");
        }
    }
  } catch (error) {
    console.error("Error in demo:", error);
  }
}

// Run the demo.
demo();