Getting started

To use Agentflow, you don't need to be a professional programmer or know any programming languages. But you should be happy working in the command line, and it helps to be able to think a little like a programmer - to have a curious and analytical mind, and enjoy solving problems.

Create an Agentflow project

To use Agentflow, you'll need a JavaScript environment installed on your machine (such as Node.js or Bun), then use your preferred package manager to bootstrap a new project. You can optionally specify the project name via additional command line options.
(Using . for the project name will scaffold the project in the current directory)

npm

npm create agentflow@latest my-agents

yarn

yarn create agentflow my-agents

bun

bun create agentflow my-agents

Follow the on-screen prompts, and once the initialiser is finished, change into the project directory and install the dependencies.

npm

cd my-agents
npm i

yarn

cd my-agents
yarn

bun

cd my-agents
bun i

Project structure

Initialising a new project results in folder structure including the following key files and folders.

• 📂 flows - The folder in which all your workflows will be created and kept.
• 📂 outputs - The folder where the results of your workflow executions will be stored.
• .env - An environment file, in which to store secrets such as AI provider API keys.
• agentflow.config.js - The configuration file for your Agentflow project.

my-agents/
├── flows/
│   └── hello-world.mdx
├── outputs/
├── .env
├── agentflow.config.js
└── package.json

Before you can run workflows, you'll need to:

• Configure your AI providers in agentflow.config.js
• Add your API keys to the .env file

Command line interface

In a project where Agentflow is installed, you can use the aflow binary to create, manage and execute workflows in your project. Use the aflow help command to see a list of available commands.

npm

npx aflow help

yarn

yarn run aflow help

bun

bunx aflow help

Learn more about the command line interface.

Project configuration

When running aflow from the command line, Agentflow will try to find and load a config file named agentflow.config.js (or .ts) from the project root. If there is also a .env file present, this is loaded first and all defined environment variables can be accessed on process.env.

import { defineConfig } from '@agentflow/core'
import { createOpenAI } from '@ai-sdk/openai'

export default defineConfig({
  providers: {
    openai: createOpenAI({
      apiKey: process.env.MY_OPENAI_KEY
    })
  }
})

Learn more about the configuring Agentflow.