Getting Started

LlamaAgents at a glance

LlamaAgents helps you build and deploy small, workflow‑driven agentic apps using LlamaIndex, locally and on LlamaCloud. Define LlamaIndex workflows, run them as durable APIs that can pause for input, optionally add a UI, and deploy to LlamaCloud in seconds.

LlamaAgents is for developers and teams building automation, internal tools, and app‑like agent experiences without heavy infrastructure work.

Build and ship small, focused agentic apps—fast. Start from either our templated LlamaIndex workflow apps, or from workflows you’ve already prototyped, iterate locally, and deploy to LlamaCloud right from your terminal in seconds.

Write LlamaIndex Python workflows, and serve them as an API. For example, make a request to process incoming files, analyze them, and return the result or forward them on to another system.
Workflow runs are durable, and can wait indefinitely for human or other external inputs before proceeding.
Optionally add a UI for user-driven applications. Make custom chat applications, data extraction and review applications.
Deploy your app in seconds to LlamaCloud. Call it as an API with your API key, or visit it secured with your LlamaCloud login.

LlamaAgents is built on top of LlamaIndex’s (soon-to-be) open source LlamaDeploy v2. LlamaDeploy is a toolchain to create, develop, and deploy workflows. The llamactl command line interface (CLI) is the main point of entry to developing LlamaDeploy applications: It can scaffold LlamaDeploy based projects with llamactl init, serve them with llamactl serve, and deploy with llamactl deployments create.

In addition to LlamaDeploy, LlamaIndex published additional SDKs to facilitate rapid development:

Our llama-cloud-services JS and Python SDKs offer a simple way to persist ad hoc Agent Data. Read more here.
Our @llamaindex/ui React library offers off-the-shelf components and hooks to facilitate developing workflow-driven UIs.

Getting Started with LlamaDeploy

LlamaDeploy is composed of the llamactl CLI for development. llamactl bootstraps an application server that manages running and persisting your workflows, and a control plane for managing cloud deployments of applications. It has some system pre-requisites that must be installed in order to work:

Make sure you have uv installed. uv is a Python package manager and build tool. llamactl integrates with it in order to quickly manage your project’s build and dependencies.
Likewise, Node.js is required for UI development. You can use your node package manager of choice (npm, pnpm, or yarn).

Install

Choose one:

Try without installing:

uvx llamactl --help

Install globally (recommended):

uv tool install -U llamactl
llamactl --help

Initialize a project

llamactl includes starter templates for both full‑stack UI apps, and headless (API only) workflows. Pick a template and customize it.

llamactl init

This will prompt for some details, and create a Python module that contains LlamaIndex workflows, plus an optional UI you can serve as a static frontend.

Application configuration is managed within your project’s pyproject.toml, where you can define Python workflow instances that should be served, environment details, and configuration for how the UI should be built. See the Deployment Config Reference for details on all configurable fields.

Develop

Once you have a project, you can run the dev server for your application:

llamactl serve

llamactl serve will

Install all required dependencies
Read the workflows configured in your app’s pyproject.toml and serve them as an API
Start up and proxy the frontend development server, so you can seamlessly write a full stack application.

For example, with the following configuration, the app will be served at http://localhost:4501/deployments/my-package. Make a POST request to /deployments/my-package/workflows/my-workflow/run to trigger the workflow in src/my_package/my_workflow.py.

[project]
name = "my-package"
# ...
[tool.llamadeploy.workflows]
my-workflow = "my_package.my_workflow:workflow"

[tool.llamadeploy.ui]
directory = "ui"

# from workflows import ...
# ...
workflow = MyWorkflow()

At this point, you can get to coding. The development server will detect changes as you save files. It will even resume in-progress workflows!

For more information about CLI flags available, see llamactl serve.

For a more detailed reference on how to define and expose workflows, see Workflows & App Server API.

Create a Deployment

LlamaDeploy applications can be rapidly deployed just by pointing to a source git repository. With the provided repository configuration, LlamaCloud will clone, build, and serve your app. It can even access GitHub private repositories by installing the LlamaDeploy GitHub app

Example:

git remote add origin https://github.com/org/repo
git add -A
git commit -m 'Set up new app'
git push -u origin main

Then, create a deployment:

llamactl deployments create

The first time you run this, you’ll be prompted to log into LlamaCloud. See llamactl auth for more info

This will open an interactive Terminal UI (TUI). You can tab through fields, or even point and click with your mouse if your terminal supports it. All required fields should be automatically detected from your environment, but can be customized:

Name: Human‑readable and URL‑safe; appears in your deployment URL
Git repository: Public HTTP or private GitHub (install the LlamaCloud GitHub app for private repos)
Git branch: Branch to pull and build from (use llamactl deployments update to roll forward). This can also be a tag or a git commit.
Secrets: Pre‑filled from your local .env; edit as needed. These cannot be read again after creation.

When you save, LlamaDeploy will verify that it has access to your repository, (and prompt you to install the GitHub app if not)

After creation, the TUI will show deployment status and logs.

You can later use llamactl deployments get to view again.
You can add secrets or change branches with llamactl deployments edit.
If you update your source repo, run llamactl deployments update to roll a new version.

Next: Read about defining and exposing workflows in Workflows & App Server API.