XML Agent
caution
This is a legacy chain, it is not recommended for use. Instead, see docs for the LCEL version.
Some language models (like Anthropic's Claude) are particularly good at reasoning/writing XML. The below example shows how to use an agent that uses XML when prompting.
import { ChatAnthropic } from "langchain/chat_models/anthropic";
import { initializeAgentExecutorWithOptions } from "langchain/agents";
import { SerpAPI } from "langchain/tools";
const model = new ChatAnthropic({ modelName: "claude-2", temperature: 0.1 });
const tools = [new SerpAPI()];
const executor = await initializeAgentExecutorWithOptions(tools, model, {
agentType: "xml",
verbose: true,
});
console.log("Loaded agent.");
const input = `What is the weather in Honolulu?`;
const result = await executor.invoke({ input });
console.log(result);
/*
https://smith.langchain.com/public/d0acd50a-f99d-4af0-ae66-9009de319fb5/r
{
output: 'The weather in Honolulu is currently 75 degrees Fahrenheit with a small craft advisory in effect. The forecast calls for generally clear skies tonight with a low of 75 degrees.'
}
*/
API Reference:
- ChatAnthropic from
langchain/chat_models/anthropic
- initializeAgentExecutorWithOptions from
langchain/agents
- SerpAPI from
langchain/tools