Skip to Content
Getting Started

Getting Started

Get Marlo running with your agent in under 5 minutes.

Create an Account

  1. Go to marshmallo.ai 
  2. Click Get Started and create your account
  3. Set up an organization
  4. Create a project within your organization

Note: Marlo is currently in beta. For early access, reach out to hello@marshmallo.ai.

Get Your API Key

  1. Navigate to your dashboard and click Create Project
    • A project represents a single agentic system
    • If you have multiple agents, create one project per system
  2. Go to Settings → Project for your project
  3. Copy the API key displayed on this page
  4. Add it to your environment:
export MARLO_API_KEY=your_api_key_here

Or add to your .env file:

MARLO_API_KEY=your_api_key_here

Install the SDK

Python

pip install marlo-sdk

TypeScript

npm install @marshmallo/marlo

Wrap Your Agent

Python

import os
import marlo
from openai import OpenAI
 
# Initialize
marlo.init(api_key=os.getenv("MARLO_API_KEY"))
marlo.instrument_openai()
 
# Register your agent
marlo.agent(
    name="my-agent",
    system_prompt="You are a helpful assistant.",
    tools=[],
    mcp=[],
)
 
# Wrap your agent logic
client = OpenAI()
 
with marlo.task(thread_id="user-123", agent="my-agent") as task:
    task.input("What is 2 + 2?")
    
    response = client.chat.completions.create(
        model="gpt-4",
        messages=[{"role": "user", "content": "What is 2 + 2?"}]
    )
    
    task.output(response.choices[0].message.content)
 
marlo.shutdown()

TypeScript

import * as marlo from '@marshmallo/marlo';
import OpenAI from 'openai';
 
// Initialize
await marlo.init(process.env.MARLO_API_KEY!);
 
// Register your agent
marlo.registerAgent(
  'my-agent',
  'You are a helpful assistant.',
  [],
  [],
);
 
// Wrap your agent logic
const client = new OpenAI();
 
const task = marlo.task('user-123', 'my-agent').start();
task.input('What is 2 + 2?');
 
const response = await client.chat.completions.create({
  model: 'gpt-4',
  messages: [{ role: 'user', content: 'What is 2 + 2?' }],
});
 
task.llm({
  model: 'gpt-4',
  usage: {
    input_tokens: response.usage?.prompt_tokens || 0,
    output_tokens: response.usage?.completion_tokens || 0,
  },
});
 
task.output(response.choices[0].message.content || '');
task.end();
 
await marlo.shutdown();

View Results

After running your agent:

  1. Go to your project in the Marlo dashboard 
  2. Click on Threads to see your conversations
  3. Click into a task to see the full trace
  4. Watch as Rewards score your agent’s performance
  5. Check Learnings to see guidance generated from your agent’s behavior

Try Without Code: Marlo Inbox

Want to see Marlo in action without integrating your own agent? Try Marlo Inbox, a complete email and calendar assistant that demonstrates the full learning loop.

Quick Start

git clone https://github.com/Marshmallo-AI/marlo-inbox.git
cd marlo-inbox
cp .env.example .env
# Fill in your credentials in .env
make install-all
make dev  # Python version
# or: make dev-ts-full  # TypeScript version

Open http://localhost:5173, sign in with Google, and start chatting. Within 10-15 interactions, you’ll see learnings appear that improve the agent’s behavior.

See the Marlo Inbox README  for detailed setup instructions.

What’s Next?

Now that you’re capturing data, explore:

Last updated on
Introduction to MarloAuto Instrumentation