The Agent executor not calling _call function

Checked other resources

[X] I added a very descriptive title to this issue.
[X] I searched the LangChain.js documentation with the integrated search.
[X] I used the GitHub search to find a similar question and didn't find it.
[X] I am sure that this is a bug in LangChain.js rather than my code.
[X] The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).

Example Code

import { ChatOpenAI } from "@langchain/openai";
import { AgentExecutor } from "langchain/agents";
import { BaseMessageChunk } from "@langchain/core/messages";
import { AgentAction, AgentFinish } from "@langchain/core/agents";
import { RunnableSequence } from "@langchain/core/runnables";
import { ChatPromptTemplate } from "@langchain/core/prompts";
import { SearxngSearch } from "@langchain/community/tools/searxng_search";

const model = new ChatOpenAI({
  maxTokens: 1000,
  model: "gpt-4",
});

// `apiBase` will be automatically parsed from .env file, set "SEARXNG_API_BASE" in .env,
const tools = [
  new SearxngSearch({
    params: {
      format: "json", // Do not change this, format other than "json" is will throw error
      engines: "google",
    },
    // Custom Headers to support rapidAPI authentication Or any instance that requires custom headers
    headers: {},
  }),
];
const prefix = ChatPromptTemplate.fromMessages([
  [
    "ai",
    "Answer the following questions as best you can. In your final answer, use a bulleted list markdown format.",
  ],
  ["human", "{input}"],
]);
// Replace this with your actual output parser.
const customOutputParser = (
  input: BaseMessageChunk
): AgentAction | AgentFinish => ({
  log: "test",
  returnValues: {
    output: input,
  },
});
// Replace this placeholder agent with your actual implementation.
const agent = RunnableSequence.from([prefix, model, customOutputParser]);
const executor = AgentExecutor.fromAgentAndTools({
  agent,
  tools,
});
console.log("Loaded agent.");
const input = `What is Langchain? Describe in 50 words`;
console.log(`Executing with input "${input}"...`);
const result = await executor.invoke({ input });
console.log(result);

Error Message and Stack Trace (if applicable)

No response

Description

The toolchain does not get to _call

 AgentExecutor.fromAgentAndTools({
        agent,
        tools: [tool],
      });

System Info

"@langchain/community": "^0.2.20",

The issue with the AgentExecutor not working for Searxng or DuckDuckgo in LangChain.js might be due to the tools not being properly integrated or recognized within the AgentExecutor. Here is an example of how to correctly instantiate and use the AgentExecutor with the SearxngSearch tool:

import { ChatOpenAI } from "@langchain/openai";
import { AgentExecutor } from "langchain/agents";
import { BaseMessageChunk } from "@langchain/core/messages";
import { AgentAction, AgentFinish } from "@langchain/core/agents";
import { RunnableSequence } from "@langchain/core/runnables";
import { ChatPromptTemplate } from "@langchain/core/prompts";
import { SearxngSearch } from "@langchain/community/tools/searxng_search";

const model = new ChatOpenAI({
  maxTokens: 1000,
  model: "gpt-4",
});

// `apiBase` will be automatically parsed from .env file, set "SEARXNG_API_BASE" in .env,
const tools = [
  new SearxngSearch({
    params: {
      format: "json", // Do not change this, format other than "json" is will throw error
      engines: "google",
    },
    // Custom Headers to support rapidAPI authentication Or any instance that requires custom headers
    headers: {},
  }),
];
const prefix = ChatPromptTemplate.fromMessages([
  [
    "ai",
    "Answer the following questions as best you can. In your final answer, use a bulleted list markdown format.",
  ],
  ["human", "{input}"],
]);
// Replace this with your actual output parser.
const customOutputParser = (
  input: BaseMessageChunk
): AgentAction | AgentFinish => ({
  log: "test",
  returnValues: {
    output: input,
  },
});
// Replace this placeholder agent with your actual implementation.
const agent = RunnableSequence.from([prefix, model, customOutputParser]);
const executor = AgentExecutor.fromAgentAndTools({
  agent,
  tools,
});
console.log("Loaded agent.");
const input = `What is Langchain? Describe in 50 words`;
console.log(`Executing with input "${input}"...`);
const result = await executor.invoke({ input });
console.log(result);
/**
 * Langchain is a framework for developing applications powered by language models, such as chatbots, Generative Question-Answering, summarization, and more. It provides a standard interface, integrations with other tools, and end-to-end chains for common applications. Langchain enables data-aware and powerful applications.
 */

Ensure that SearxngSearch is correctly implemented and extends the Tool class, and that it is properly recognized by the AgentExecutor. If these tools are not correctly implemented or recognized, the AgentExecutor will not be able to use them, leading to the issue you are experiencing ^[1]^[2].

To continue talking to Dosu, mention @dosu.

What is the update on this?

The issue with the AgentExecutor not calling the _call function in your setup might be due to the customOutputParser not correctly parsing the output into a format that the AgentExecutor can use. Ensure that the customOutputParser function returns an object that matches the AgentAction or AgentFinish interface.

Here's an example of how to correctly implement the customOutputParser:

const customOutputParser = (text: string): AgentAction | AgentFinish => {
  if (text.includes("Final Answer:")) {
    const parts = text.split("Final Answer:");
    const input = parts[parts.length - 1].trim();
    const finalAnswers = { output: input };
    return { log: text, returnValues: finalAnswers };
  }
  const match = /Action: (.*)\nAction Input: (.*)/s.exec(text);
  if (!match) {
    throw new Error(`Could not parse LLM output: ${text}`);
  }
  return {
    tool: match[1].trim(),
    toolInput: match[2].trim().replace(/^"+|"+$/g, ""),
    log: text,
  };
};

Ensure that the SearxngSearch tool is correctly implemented and extends the Tool class. The AgentExecutor should be able to recognize and use it if it is properly configured.

If the issue persists, you might want to check the verbose logs to get more insights into the execution process. This can help identify any problems with tool invocation or output parsing.

For more details on how the SearxngSearch tool handles API requests and responses, you can refer to the implementation in the LangChain.js repository ^[1]^[2].

To continue talking to Dosu, mention @dosu.

langchain-ai / langchainjs