How to build a Claude-powered CLI chatbot in 30 lines of Node.js

Most Claude tutorials show you how to build a web app. But sometimes you just want to ask AI a question from your terminal — without opening a browser, without a UI, without anything except a prompt.

Here's how to build a fully functional Claude CLI chatbot with conversation memory in under 30 lines of Node.js.

What you'll build

• Interactive REPL loop in the terminal
• Persistent conversation memory (Claude remembers the context)
• Graceful exit on quit or Ctrl+C
• Works with any Claude model

Prerequisites

npm install @anthropic-ai/sdk readline

You'll need a Claude API key. If you don't have one, you can use a flat-rate wrapper like SimplyLouie that gives you Claude access for $2/month.

The full code

// cli-chat.js
import Anthropic from '@anthropic-ai/sdk';
import readline from 'readline';

const client = new Anthropic({ apiKey: process.env.CLAUDE_API_KEY });
const history = [];

const rl = readline.createInterface({
  input: process.stdin,
  output: process.stdout
});

function prompt() {
  rl.question('You: ', async (input) => {
    if (input.toLowerCase() === 'quit') {
      console.log('Bye!');
      rl.close();
      return;
    }

    history.push({ role: 'user', content: input });

    const response = await client.messages.create({
      model: 'claude-3-5-haiku-20241022',
      max_tokens: 1024,
      messages: history
    });

    const reply = response.content[0].text;
    history.push({ role: 'assistant', content: reply });

    console.log(`Claude: ${reply}\n`);
    prompt();
  });
}

console.log('Claude CLI — type "quit" to exit\n');
prompt();

Run it

export CLAUDE_API_KEY=your_key_here
node cli-chat.js

Output:

Claude CLI — type "quit" to exit

You: what's the capital of France?
Claude: Paris.

You: what language do they speak there?
Claude: French. (Notice how it remembers we were talking about France.)

You: quit
Bye!

Add streaming so it feels instant

The version above waits for the full response before printing. Here's how to stream tokens as they arrive:

async function streamReply(history) {
  process.stdout.write('Claude: ');

  const stream = await client.messages.stream({
    model: 'claude-3-5-haiku-20241022',
    max_tokens: 1024,
    messages: history
  });

  let fullText = '';

  for await (const chunk of stream) {
    if (chunk.type === 'content_block_delta' && chunk.delta.type === 'text_delta') {
      process.stdout.write(chunk.delta.text);
      fullText += chunk.delta.text;
    }
  }

  console.log('\n');
  return fullText;
}

Replace the client.messages.create call with:

const reply = await streamReply(history);
history.push({ role: 'assistant', content: reply });

Now tokens print as they're generated — the same UX as Claude.ai in your terminal.

Keep conversation history from growing forever

If you chat for a long time, the history array grows and you'll eventually hit token limits. Trim it:

function trimHistory(history, maxTurns = 10) {
  // Keep last N turns (each turn = 2 messages: user + assistant)
  return history.slice(-(maxTurns * 2));
}

Call before each API request:

const trimmed = trimHistory(history);
const response = await client.messages.create({
  model: 'claude-3-5-haiku-20241022',
  max_tokens: 1024,
  messages: trimmed
});

Add a system prompt to give Claude a persona

const response = await client.messages.create({
  model: 'claude-3-5-haiku-20241022',
  max_tokens: 1024,
  system: 'You are a senior software engineer. Be concise. Give code examples when relevant.',
  messages: trimmed
});

Now your CLI chatbot acts as a coding assistant, not a general-purpose assistant.

Package it as a global CLI command

Add to package.json:

{
  "name": "my-claude-cli",
  "version": "1.0.0",
  "type": "module",
  "bin": {
    "ask": "./cli-chat.js"
  }
}

Add shebang to top of cli-chat.js:

#!/usr/bin/env node

Install globally:

npm install -g .

Now you can run ask from anywhere on your system.

Cost

Claude 3.5 Haiku costs $0.80/million input tokens and $4/million output tokens. A typical CLI conversation is ~500 tokens — that's fractions of a cent per session.

If you want zero-math flat-rate access, SimplyLouie wraps the Claude API for $2/month with a simple REST endpoint — no API key management, no usage tracking, just POST and get a response.

What's next?

• Add a --file flag to pipe file contents into the conversation
• Pipe stdin: cat error.log | ask 'what does this error mean?'
• Save conversation history to a JSON file for later review

What would you add to this CLI? Drop it in the comments.