LlamaIndex Workflows
OrchestrationMIT
LlamaIndex Workflows is a Python library for controlling the execution flow of an AI application. You split the application into steps; each step receives an event, does its work and returns an event, which decides the next step. Events are Pydantic objects, a context object holds per-run state, and a handler streams events while a run is in progress. Developers install it on its own with pip and can serve workflows with the llamactl command line tool.
The code lives in the run-llama/llama-agents repository. The durable workflows page says resumption is at-least-once, so step side effects must be safe to repeat.

Pinout
- MCP server: not checked yet
- HTTP API: not checked yet
- command line: supported, verified
- Python SDK: supported, verified
- TypeScript SDK: not checked yet
- OpenAPI description: not checked yet
| Interface | Status | Source we read | Read on |
|---|---|---|---|
| MCP server | Not checked yet | — | — |
| HTTP API | Not checked yet | — | — |
| Command line | Supported, verified | Repository README, llamactl commands | |
| Python SDK | Supported, verified | Workflows documentation, install line (pip install llama-index-workflows) | |
| TypeScript SDK | Not checked yet | — | — |
| OpenAPI description | Not checked yet | — | — |
Facts
| License | MIT. GitHub repository page, Resources, read . |
|---|---|
| Repository name | run-llama/llama-agents. GitHub repository page, read . |
| Python package | pip install llama-index-workflows (standalone install). Workflows documentation, read . |
| Programming model | Event-driven, step-based control of execution flow; each step receives an event and returns an event. Workflows documentation, read . |
| Events | User-defined Pydantic objects, plus built-in start and stop events. Workflows documentation, read . |
| Streaming | run() returns a handler; stream_events() yields events while the workflow runs. Workflows documentation, read . |
| Run state | Per-run shared state through ctx.store. Workflows documentation, read . |
| Persistence | No built-in checkpointer; state is saved with Context.to_dict() and restored with Context.from_dict(). Durable workflows page, read . |
| Runtime plugin | DBOS runtime journals step transitions to a database so a crashed workflow can resume. Durable workflows page, read . |
| Command line tool | uv tool install llamactl, then llamactl init, serve and deployments create. GitHub repository README, read . |
| Latest package release | llama-index-workflows@2.26.1. GitHub Releases page, read . |
| Last commit | . GitHub API, latest commit on the default branch of run-llama/llama-agents, read . |
Call example
class HelloWorkflow(Workflow):
@step
async def greet(self, ev: StartEvent) -> StopEvent:
return StopEvent(result=f"Hello, {ev.name}")Limits
- There is no built-in checkpointer; persistence is manual or through a runtime plugin.
- Resume is at-least-once, so step side effects must be safe to repeat.
Strengths
- Installs standalone with pip install llama-index-workflows.
- Events are typed Pydantic objects and runs can stream events.
- The repository is MIT licensed.
Alternatives to LlamaIndex Workflows
All orchestration tools for agents
Low-level framework for stateful, long-running agent workflows built as graphs.

Typed Python SDK for building agents across many model providers.

Durable execution platform, open source, used to run long-lived agents.

Open-source Python framework for multi-agent workflows; separate commercial platform.
Last check: . Method: public repository and docs read automatically, page and day written down. Next check due within 90 days. How we check
Is this your tool? Write to hello@agent-tools.directory from an address on its domain to correct a pin or a fact.