burr: a state-machine framework for developing and monitoring stateful AI agents and decision-making applications

What it solves

Apache Burr simplifies the development of applications that require complex decision-making and state management, such as AI agents, chatbots, and simulations. It provides a structured way to manage state and track decisions without forcing a specific LLM or data management strategy on the developer.

How it works

Burr allows developers to express their application as a state machine (a graph or flowchart) using simple Python functions. These functions, called actions, are tied together by transitions to define the flow of the application. The framework includes a pluggable persistence layer to save and load application state and a dedicated UI for real-time telemetry, monitoring, and debugging of the state machine's execution.

Who it’s for

Python developers building stateful AI applications, AI agents, or non-LLM simulations and workflows that require explicit state tracking and idempotent workflows.

Highlights

• Framework-Agnostic: Works with any LLM or existing AI framework (e.g., LangChain, LlamaIndex).
• Real-time Telemetry UI: Includes a built-in UI to visualize and trace execution for easier debugging.
• State Management: Provides a low-abstraction library for building and managing state machines with simple Python functions.
• Pluggable Persisters: Supports saving and loading application state for memory and persistence.
• Versatile Use-Cases: Applicable to both LLM-based apps (like RAG chatbots) and non-LLM tasks (like hyperparameter tuning).