LlamaAgents Builder: From Prompt to Implemented AI Agent in Minutes

In this article, you’ll learn how to build, deploy, and test a zero-code AI agent with the LlamaAgents Builder at LlamaCloud.

Topics we will cover include:

• How to create a document classification agent using a natural language command.
• How to deploy an agent to a GitHub-backed app without writing code.
• How to check the agent used on invoices and contracts in the LlamaCloud interface.

Let’s not waste any more time.

Introduction

Creating an AI agent for tasks like analyzing and processing documents used to be automated would require endless hours of configuration, coding, and deployment battles. Until now.

This article covers the process of building, deploying, and deploying an intelligent agent from scratch without writing a single line of code, using LlamaAgents Builder. Even better, we will treat it as an application in a software environment that will be 100% ours.

We will complete the entire process in a few minutes, so time is of the essence: let’s get started.

Building with LlamaAgents Builder

LlamaAgents Builder is one of the new features in LlamaCloud web platform, whose flagship product was initially introduced as LlamaParse. A bit of a confusing mix of words, I know! In the meantime, keep in mind that we will access the agent builder through this link.

The first thing you should see is a home menu like the one shown in the screenshot below. If this is not what you see, try clicking the “LlamaParse” icon in the upper left corner, and you should see this — at least at the time of writing.

LlamaParse home menu

Note that, in this example, we are running under a newly created free plan account, which allows up to 10,000 pages to process.

See the “Agents” block at the bottom right? This is where the LlamaAgents Builder resides. Although it’s in beta at the time of writing, we can already create agent-based workflows, as we’ll see.

Once we click on it, a new screen will open with chat like Gemini, ChatGPT, and others. You’ll find several suggested workflows for what you’d like your agent to do, but we’ll specify our own by typing the following command into the input box below. Just natural language, no code at all:

Create an agent that separates documents into “Contracts” and “Invoices”. For contracts, exclude the signing parties; on invoices, total amount and date.

It specifies what the agent should do with a natural language command

Just send the information, and the magic will begin. With an incredible level of transparency in the consultation process, you will see the steps completed and the progress made so far:

AgentBuilder creates the workflow for our agent

After a few minutes, the creation process will be completed. You can not only see the full diagram of the workflow, which has grown gradually throughout the process, but you also get a short and clear explanation of how to use your newly created agent. Simply amazing.

Agent workflow is created

The next step is to send our agent to be used. In the upper right corner, you can see “Push & Use” button. This starts the process of publishing your agent’s workflow software packages to the GitHub repository, so make sure you have a registered account on GitHub first. You can easily register with an existing Google or Microsoft account, for example. Once you connect the LlamaCloud platform to your GitHub account, it’s very easy to push and use your agent, which specifies that your agent, just gives you a password: it:

Pushes and releases agent workflows to GitHub

The process will take a few minutes, and you will see a series of messages similar to the command line from the fly. Once completed and your agent status appears as “Running“, you will see the last few messages like this:

“Uvicorn” messages indicate that our agent has been deployed and is running as a microservice API within the LlamaCloud infrastructure. If you’re familiar with FastAPI endpoints, you might want to try it programmatically using the API, but in this tutorial, we’ll keep things simple (we promised zero coding, right?) and try everything ourselves in the LlamaCloud user interface.

To do this, click “Visit” button from the top:

The agent used is also effective

Now comes the most exciting part. You should have been taken to a playground page called “Review,” where you can try out your agent. Start by uploading a file, for example, a PDF document containing an invoice or contract. If you don’t have one, just create a mock example document yourself using Microsoft Word, Google Docs, or a similar tool, like this one:

LlamaCloud Agent Testing UI: processing an invoice

As soon as the document is uploaded, the agent starts working on its own, and after a few seconds, it will parse your document and extract the necessary data fields, depending on the type of document. You can see this result in the right panel in the image above: the total amount and invoice date have been correctly issued by the agent.

How about uploading a sample document containing the contract now?

LlamaCloud Agent Testing UI: processing the contract

As expected, the document is now considered a contract, and in this instance, the extracted information contains the names of the signing parties.

Well done! As you continue to use examples, make sure you approve or reject them based on whether they have been processed correctly: this helps the agent learn from the feedback.

Test cases for agents and their status

Wrapping up

We have seen how to create and use, step by step and without lines of code, an AI agent that can classify documents and process them in different ways depending on the type of document – all in a few minutes and within the new feature of LlamaCloud, LlamaAgents Builder.