Apache Burr

Apache Burr (incubating) is a Python framework designed for building applications that make decisions - chatbots, agents, simulations, and complex workflows. By expressing applications as state machines with explicit state transitions, Burr provides clarity and observability that traditional frameworks lack.

The core concept revolves around actions that read from and write to application state, creating a clear flow of data and decisions. Each action is a Python function decorated with @action that specifies which state variables it reads and writes. This explicit contract makes applications easier to understand, test, and maintain.

Burr's built-in telemetry UI provides real-time visualization of state machine execution, showing how state evolves over time and which decisions led to specific outcomes. This observability is crucial for debugging complex AI applications where understanding the decision flow is often more important than just seeing the final output.

The framework includes pluggable persisters that can save and restore application state from various backends (memory, file system, databases), enabling applications to resume from any point and supporting patterns like human-in-the-loop workflows, error recovery, and batch processing.

Integrations with popular libraries and frameworks make Burr flexible without being opinionated. You can use any LLM provider, integrate with existing Python libraries, and build custom actions that delegate to specialized tools while maintaining the benefits of state machine clarity and observability.

Burr works well for both LLM and non-LLM use cases, including time-series forecasting, hyperparameter tuning, and any application requiring stateful workflows. Its dependency-free core and simple API make it suitable for everything from research prototypes to production systems.