Overview

LlamaAgents at a Glance

LlamaAgents is the most advanced way to build agent workflows. Author and run multi-step document agents with our open-source Agent Workflows, then deploy them to LlamaCloud or your own infrastructure.

Stitch together Parse, Extract, Split, Classify, and custom operations into Workflows that perform knowledge tasks on your documents. When you need full control, it’s real Python underneath: fork and extend without a rewrite. Agent Workflows give you event-driven orchestration with branching, parallelism, human-in-the-loop review, durability, and observability.

Get Started

• Build locally: Use the llamactl CLI to create projects from starter templates, develop and serve workflows on your machine, then deploy to LlamaCloud or self-host. You can also use Agent Workflows directly in your own Python applications—run them as async processes or mount them as endpoints in your existing server.

• Deploy a starter template: Choose a template in LlamaCloud and deploy it without using the command line.

• Go deeper: Combine local development with cloud services. Use Agent Workflows for orchestration and WorkflowClient to call deployed workflows via REST or the typed Python client.

Components

llamactl CLI: Development and deployment for local workflow apps. Initialize from starter templates, serve locally, and deploy to LlamaCloud or export for self-hosting.

Agent Workflows: The event-driven orchestration framework at the core. Use it as an async library in your own code, or let llamactl serve it. Built-in durability and observability.

llama-cloud-services: LlamaCloud document primitives (Parse, Extract, Classify), Agent Data for structured storage, and vector indexes. llamactl handles authentication when deploying to the cloud.

@llamaindex/ui: React hooks for workflow-powered frontends. Deploy alongside your backend with llamactl.

Workflows Client: Call deployed workflows via REST API or typed Python client.