Apache Burr

Apache Burr (currently incubating at the ASF) is a Python framework that models LLM applications as state machines — explicit Actions, State, and Transitions instead of the chained-call abstractions favored by LangChain or the graph-of-nodes pattern in LangGraph. Burr's pitch is operational: state management, replay, debugging, and persistence are framework primitives rather than escape hatches, which is why production teams cite it specifically for getting from prototype to deployed. The cookbook covers concrete patterns: counter and 'choose your own adventure' toys; GPT-like multimodal chatbots; conversational RAG (often paired with Hamilton); divide-and-conquer agent patterns; web-server deployment; provisioning and monitoring; guardrails; hyperparameter tuning; simulations. The framework ships a bundled Burr UI for post-hoc state-machine visualization and execution trace monitoring — debugging surface that Reddit's LocalLlama community and engineers at Peanut Robotics, Watto.ai, Paxton AI, and Provectus specifically called out in public testimonials. Integrations include Hamilton, Streamlit, OpenTelemetry, Traceloop, Langchain, Pydantic, Haystack, and Ray. Burr is free and open-source under the Apache 2.0 license, with optional paid hosting from DAGWorks for teams that don't want to run telemetry infrastructure themselves.