Getting Started

The easiest way to start building agents with Open Agents Builder is to create a free, cloud-hosted instance at openagentsbuilder.com. Once that’s done, you can start building agents right away with full API access.

How Does It Work?

Building AI Agents Made Easy 🤖 – Watch Video

Local Setup

The cloud edition is great for getting started, building PoCs, or if you just want to avoid technical hassles and begin quickly. However, if you plan to build custom integrations, use open-source LLM models, or develop custom AI tools, you’ll likely need to deploy the software locally.

Fortunately, it only takes a few shell commands!

First, check out the project:

git clone https://github.com/CatchTheTornado/open-agents-builder
cd open-agents-builder

Make sure you have Node.js version 22 installed, which is required by the project. If not, you can switch to version 22 using nvm (see how to install nvm if you haven’t already):

nvm install 22
nvm use 22

If you plan to use the Attachments module to extract text from PDF and Office files, you’ll need to install some Python dependencies. If you don’t already have a local Python environment set up, the easiest approach is to use pipx instead of pip, so the packages are installed into an isolated, global environment:

Install pipx using brew install pipx (on macOS) or apt-get install pipx (on Debian), and then run:

pipx install markitdown
pipx install poppler

Next, edit the environment variables:

cp .env.defaults .env.local

Then open and modify the .env.local file:

LLM_PROVIDER=openai
LLM_MODEL=gpt-4o

# OpenAI API key
OPENAI_API_KEY=

# Resend.com API key
RESEND_API_KEY=

# Absolute app URL starting with http://
NEXT_PUBLIC_APP_URL=http://localhost:3000
APP_URL=http://localhost:3000
PORT=3000 # 3002 is the production app port
NEXT_PUBLIC_EMAIL_FROM=no-reply@.... # According to your Resend.com settings
NEXT_PUBLIC_ENV=dev

OpenAI API Key and LLM Support

We use Vercel AI to access LLMs and embedding models, which allows you to switch the LLM provider to any supported by the Vercel AI SDK (e.g., Google Gemini, Claude, etc.).

Currently, due to extensive usage of Tools and Structured Outputs, the application works best with OpenAI models.

Go to platform.openai.com to create your API key, and then set it in the environment variables shown above.

Resend API Key

We use Resend.com for sending emails. You can create an account and an API key on resend.com.

This step isn’t mandatory for testing and extending the app, but without it, email sending won’t work.

Running the Application

yarn
yarn dev

That’s it! Now open your browser and navigate to http://localhost:3000.

Note on the Tech Stack

As you may have noticed, Open Agents Builder requires almost no external dependencies. We use the SQLite database—a single instance per data owner (account)—capable of enormous data storage (up to 281 TB). It should easily handle a few terabytes of data.

This approach provides minimal dependencies and enhanced security (each SQLite file is encrypted and isolated, which is beneficial for multi-tenant setups).

Porting Open Agents Builder to another database (e.g., Postgres) shouldn’t require major effort. All data operations are done via Drizzle and follow a Repository design pattern.