Capability · Framework — orchestration

Burr

Burr treats an LLM app as a finite state machine — each Action is a node that reads state, runs logic, and writes state. That makes the flow explicit, testable, and resumable: the same code runs in dev, in prod, and inside the Burr UI that renders the graph and traces state transitions. It's a lightweight alternative to LangGraph for teams who prefer pure-Python ergonomics.

Framework facts

Category: orchestration
Language: Python
License: BSD-3-Clause
Repository: https://github.com/dagworks-inc/burr

Install

pip install 'burr[start]'
burr   # launch the local UI at http://localhost:7241

Quickstart

from burr.core import ApplicationBuilder, State, action

@action(reads=['prompt'], writes=['response'])
def answer(state: State) -> State:
    return state.update(response=f'echo: {state["prompt"]}')

app = (
    ApplicationBuilder()
    .with_actions(answer)
    .with_transitions(('answer', 'answer'))
    .with_entrypoint('answer')
    .with_state(prompt='hello')
    .build()
)
action_, result, state = app.run(halt_after=['answer'])
print(state['response'])

Alternatives

LangGraph — larger ecosystem, LangChain-tied
Prefect — general workflow engine with LLM support
Hamilton — DAGWorks sibling for dataflow
Temporal — durable workflow alternative

Frequently asked questions

Burr vs LangGraph?

Both model LLM apps as graphs/state machines. Burr is pure Python with a lighter abstraction and a polished local UI; LangGraph is tied to the LangChain ecosystem with deeper built-in agent primitives. Choose based on how invested you already are in LangChain.

Does Burr replace my orchestration framework?

Usually not — it complements them. Burr owns the 'in-request state machine', while something like Prefect or Airflow owns 'scheduled pipelines'. They compose well.

Sources

Burr — docs — accessed 2026-04-20
Burr GitHub — accessed 2026-04-20