Apache Burr

Apache Burr (Incubating) is a free, open-source Python framework in the AI development frameworks category that models applications as explicit state machines. Licensed under Apache 2.0 with no usage limits or gated features in the core framework, it provides built-in observability, debugging, and persistence for AI agent workflows, chatbots, and multi-step pipelines.

Originally created by DAGWorks Inc. and now incubating at the Apache Software Foundation, Burr takes a fundamentally different approach to AI orchestration compared to chain-based or graph-based frameworks like LangChain and LangGraph. Instead of implicit data flows, every application step is a defined action with typed state reads and writes, connected by explicit conditional transitions. This state-machine paradigm makes complex agent behaviors—including loops, branches, retries, and human-in-the-loop checkpoints—first-class citizens that are visible, testable, and reproducible.

The framework's core API uses Python decorators to define actions and a builder pattern to wire them into applications. Developers write standard Python functions decorated with @action, specifying which state keys each action reads and writes. Transitions between actions can be conditional, enabling dynamic routing based on runtime state. This means applications are testable with standard pytest, debuggable with standard Python debuggers, and readable without learning a custom DSL or YAML configuration. The GitHub repository (github.com/DAGWorks-Inc/burr) has accumulated approximately 2,500 stars and contributions from over 40 developers, reflecting steady community growth since the project's initial release in late 2023.

Burr ships with a bundled local telemetry UI at no additional cost, providing step-by-step execution traces, state inspection at every transition, and the ability to replay executions from any checkpoint. Unlike competing observability solutions such as LangSmith, which requires a separate paid subscription, Burr's UI runs entirely locally with no external accounts or API keys required.

Persistence is handled through a pluggable backend system supporting in-memory, SQLite, PostgreSQL, Redis, and custom implementations. This enables long-running workflows to checkpoint state and resume after failures or server restarts, making Burr suitable for production deployments where reliability is non-negotiable. Built-in FastAPI integration allows applications to be exposed as REST APIs with minimal boilerplate, and the framework supports both synchronous and asynchronous execution patterns along with streaming responses.

The project entered the Apache Software Foundation Incubator in 2025, transitioning governance from DAGWorks Inc. to a vendor-neutral community model. This incubation status brings structured IP clearance, transparent release processes, and a defined path toward graduation as a top-level Apache project. For teams evaluating the framework, the primary trade-off is ecosystem breadth versus architectural clarity: LangChain offers significantly more pre-built integrations and community extensions, while Burr provides stronger guarantees around state visibility, reproducibility, and debuggability—qualities particularly valued in regulated industries and enterprise environments that require auditable AI workflows.