Apache Burr: Build reliable AI agents with simple Python API
Pure Python framework for chatbots to multi-agent systems with built-in observability.
Apache Burr (Incubating) is a pure Python framework designed to make building reliable AI agents and applications straightforward. It provides a clean, composable API using Python functions and decorators—no DSL or YAML required. Developers define applications as sets of actions and transitions, enabling everything from simple chatbots to complex multi-agent systems. Key features include built-in observability with a real-time UI for monitoring state changes and debugging, automatic state persistence to disk or databases, human-in-the-loop capabilities for approval workflows, support for branching and parallel execution (fan-out/fan-in), and robust testing and replay functionality. This makes Burr particularly suited for production-grade AI systems where reliability and debugging are critical.
Burr integrates seamlessly with popular tools and frameworks such as OpenAI, Anthropic, LangChain, Hamilton, Streamlit, FastAPI, Haystack, Instructor, Pydantic, and PostgreSQL, ensuring no lock-in. The framework is already trusted by engineers at companies like Peanut Robotics (used for AI-driven robot decision-making), Watto.ai (for modular AI applications), Paxton AI (for state snapshots and debugging), Provectus, CognitiveGraphs, and TaskHuman. Users highlight its ease of use, robust state management, and powerful UI for debugging compared to alternatives like LangChain, CrewAI, and AutoGen. Burr is an Apache Incubating project, indicating strong community governance and long-term viability.
- Pure Python API with decorators – no DSL or YAML, just Python functions and transitions.
- Built-in observability UI for real-time state monitoring, debugging, and replay of past runs.
- Supports human-in-the-loop workflows, parallel execution (fan-out/fan-in), and integrations with OpenAI, Anthropic, LangChain, FastAPI, and more.
Why It Matters
Apache Burr offers a production-ready, observable alternative to complex agent frameworks, reducing debugging time and enabling reliable AI decision-making at scale.