Configuration

Deep agents are highly configurable. This page covers the main configuration options for customizing agent behavior.

Learn more: For comprehensive customization guides, see the Customization documentation.

Model Configuration

By default, deep agents use claude-sonnet-4-5-20250929. You can use any LangChain-compatible chat model.

import { ChatAnthropic } from "@langchain/anthropic";
import { ChatOpenAI } from "@langchain/openai";
import { createDeepAgent } from "deepagents";

// Using Anthropic
const agent = createDeepAgent({
  model: new ChatAnthropic({
    model: "claude-sonnet-4-20250514",
    temperature: 0,
  }),
});

// Using OpenAI
const agent2 = createDeepAgent({
  model: new ChatOpenAI({
    model: "gpt-5",
    temperature: 0,
  }),
});

System Prompt

Deep agents include a comprehensive built-in system prompt. Add use-case specific instructions to guide agent behavior.

import { createDeepAgent } from "deepagents";

const researchInstructions = `You are an expert researcher. 
Your job is to conduct thorough research, and then write a polished report.`;

const agent = createDeepAgent({
  systemPrompt: researchInstructions,
});

Interrupt Configuration (Human-in-the-Loop)

Configure which tools require human approval before execution.

Learn more: See Human-in-the-Loop for detailed patterns.

import { createDeepAgent } from "deepagents";

const agent = createDeepAgent({
  tools: [getWeather],
  interruptOn: {
    get_weather: {
      allowedDecisions: ["approve", "edit", "reject"],
    },
  },
});

API Reference